nasa · pdf file · 2013-09-06identification and quantitative mapping of surface...
TRANSCRIPT
JPL Publication 93-26, Vol. 3
Summaries of the Fourth Annual JPLAirborne Geoscience WorkshopOctober 25-29, 1993
Volume 3. AIRSAR Workshop
Jakob van ZylEditor
October 25, 1993
NASANational Aeronautics and
Space Administration
Jet Propulsion LaboratoryCalifornia Institute of Technology
Pasadena, Californta
https://ntrs.nasa.gov/search.jsp?R=19950017518 2018-05-16T13:52:48+00:00Z
This publication was prepared by the Jet Propulsion Laboratory, California
Institute of Technology, under a contract with the National Aeronautics and
Space Administration.
ABSTRACT
This publication contains the summaries for the Fourth Annual JPL Airborne
Geoscience Workshop, held in Washington, D. C. on October 25-29, 1993. The
main workshop is divided into three smaller workshops as follows:
The Airborne Visible/Infrared h-naging Spectrometer (AVIRIS)
workshop, on October 25-26. The summaries for this workshop
appear ill Volume 1.
The Thermal Infrared Multispectral Scanner (T1MS) workshop, on
October 27. The summaries for this workshop appear in Volume 2.
The Airborne Synthetic Aperture Radar (AIRSAR) workshop, on
October 28-29. The summaries for this workshop appear in
Volume 3.
iii
FOREWORD
In the text and the figure captions of some of the papers for the AVIRIS and
AIRSAR Workshops, reference is made to color slides. A pocket containing a setof all the slides mentioned in those papers appears at the end of this volume.
iv
CONTENTS
Volume 1: AVIRIS Workshop
Use of Spectral Analogy to Evaluate Canopy Reflectance Sensitivity to
Leaf Optical Property ..................................................................................................... 1Frdd&ic Baret, Vern C. Vanderbilt, Michael D. Steven, and
Stephane Jacquemoud
A Tool for Manual Endmember Selection and Spectral Unmixing ....................... 3
C. Ann Bateson and Brian Curtiss
Lithologic Discrimination and Alteration Mapping From AVIRIS Data,Socorro, New Mexico ..................................................................................................... 7
K. K. Beratan, N. DeLillo, A. Jacobson, R. Biota, mid C. E. Chapin
Automating Spectral Unmixing of AVIRIS Data Using Convex Geometry
Concepts ......................................................................................................................... 11
Joseph W. Boardman
AVIRIS Calibration Using the Cloud-Shadow Method ........................................... 15
K. L. Carder, P. Reinersman, and R. F. Chert
Field Observations Using an AOTF Polarimetric Imaging Spectrometer ............. 19
Li-Jen Cheng, Mike Hamilton, Colin Mahoney, and
George Reyes
Instantaneous Field of View and Spatial Sampling of the Airborne
Visible/Infrared Imaging Spectrometer (AVIRIS) ................................................... 23Thomas G. Chrien and Robert O. Green
Airborne Visible/Infrared Imaging Spectrometer (AVIRIS): Recent
Improvements to the Sensor ........................................................................................ 27Thomas G. Chrien, Robert O. Green, Charles M. Sarture,
Christopher Chovit, Michael L. Eastwood, and
Bjorn T. Eng
Comparison of Three Methods for Materials Identification and Mapping
with Imaging Spectroscopy ......................................................................................... 31
Roger N. Clark, Gregg Swayze, Joe Boardman, and Fred Kruse
Comparison of Methods for Calibrating AVIRIS Data to GroundReflectance ...................................................................................................................... 35
Roger N. Clark, Gregg Swayze, Kathy Heidebrecht,Alexander F. H. Goetz, and Robert 0. Green
The U.S. Geological Survey, Digital Spectral Reflectance Library: Version
1:0.2 to 3.0 _m. .............................................................................................................. 37
Roger N. Clark, Gregg A. Swayze, Trude V. V. King,
Andrea J. Gallagher, and Wendy M. Calvin
Application of a Two-Stream Radiative Transfer Model for Leaf Lignin
and Cellulose Concentrations From Spectral Reflectance Measurements(Part 1) ............................................................................................................................. 39
James E. Conel, Jeannette van den Bosch, and Cindy I. Grove
Application of a Two-Stream Radiative Transfer Model for Leaf Lignin
and Cellulose Concentrations From Spectral Reflectance Measurements(Part 2) ............................................................................................................................. 45
James E. Conel, Jeannette van den Bosch, and Cindy I. Grove
Discrimination of Poorly Exposed Lithologies in AVIRIS Data ............................. 53
William H. Farrand and Joseph C. Harsanyi
Spectral Decomposition of AVIRIS Data ................................................................... 57
Lisa Gaddis, Laurence Soderblom, Hugh Kieffer, Kris Becket,Jim Torson, and Kevin Mullins
Remote Sensing of Smoke, Clouds, and Fire Using AVIRIS Data ......................... 61
Bo-Cai Gao, Yoram J. Kaufinan, and Robert O. Green
Use of Data from the AVIRIS Onboard Calibrator ................................................... 65Robert 0. Green
Inflight Calibration of AVIRIS in 1992 and 1993 ...................................................... 69
Robert O. Green, James E. Conel, Mark Hehnlinger,Jeannette van den Bosch, Chris Chovit, and Toni Chrien
Estimation of Aerosol Optical Depth and Additional Atmospheric
Parameters for the Calculation of Apparent Reflectance from Radiance
Measured by the Airborne Visible/Infrared Imaging Spectrometer ..................... 73Robert O. Green, James E. Conel, and Dar A. Roberts
Use of the Airborne Visible/Infrared Imaging Spectrometer to
Calibrate the Optical Sensor on board the Japanese Earth
Resources Satellite-1 ...................................................................................................... 77
Robert O. Green, James E. Conel, Jeannette van den Bosch,
and Masanobu Shimada
A Proposed Update to the Solar Irradiance Spectrum Used in LOWTRAN
and MODTRAN ............................................................................................................ 81
Robert 0. Green and Bo-Cai Gao
vi
A Role for AVIRIS in the Landsat and Advanced Land Remote Sensing
System Program ............................................................................................................. 85Robert O. Green and John J. Simmonds
Estimating Dry Grass Residues Using Landscape Integration Analysis .............. 89
Quinn J. Hart, Susan L. Ustin, Lian Duan, and George Scheer
Classification of High Dimensional Multispectral Image Data .............................. 93
Joseph P. Hoffbeck and David A. Landgrebe
Simulation of Landsat Thematic Mapper Imagery Using AVIRIS
Hyperspectral Imagery ................................................................................................. 97Linda S. Kahnan and Gerard R. Peltzer
The Effects of AVIRIS Atmospheric Calibration Methodology on
Identification and Quantitative Mapping of Surface Mineralogy,
Drum Mtns., Utah ......................................................................................................... 101
Fred A. Kruse and John L. Dwyer
AVIRIS Spectra Correlated With the Chlorophyll Concentration of a Forest
Canopy ............................................................................................................................ 105
John Kupiec, Geoffrey M. Smith, and Paul J. Curran
AVIRIS and TIMS Data Processing and Distribution at the Land Processes
Distributed Active Archive Center ............................................................................. 109
G. R. Mah and J. Myers
Measurements of Canopy Chemistry With 1992 AVIRIS Data atBlackhawk Island and Harvard Forest. ...................................................................... 113
Mary E. Martin and John D. Aber
Classification of the LCVF AVIRIS Test Site With a Kohonen
Artificial Neural Network ............................................................................................ 117
Erzse'bet Merdnyi, Robert B. Singer, and William H. Farrand
Extraction of Auxiliary Data from AVIRIS Distribution Tape for Spectral,
Radiometric, and Geometric Quality Assessment .................................................... 121
Peter Meyer, Robert 0. Green, and Thomas G. Chrien
Preprocessing: Geocoding of AVIRIS Data using Navigation, Engineering,DEM, and Radar Tracking System Data .................................................................... 127
Peter Meyer, Steven A. Larson, Earl G. Hansen, and Klaus I. Itten
vii
The Data Facility of the Airborne Visible/Infrared Imaging Spectrometer(AVIRIS) .......................................................................................................................... 133
Pia I. Nielsen, Robert O. Green, Alex T. Murray, Bjorn T. Eng,H. Ian Novack, Manuel Solis, and Martin Olah
Mapping of the Ronda Peridotite Massif (Spain) From AVIRIS Spectro-
Imaging Survey: A First Attempt ................................................................................ 137P. C. Pinet, S. Chabrillat, and G. Ceuleneer
Spectral Variations in a Collection of AVIRIS Imagery ........................................... 141
John C. Price
Monitoring Land Use and Degradation Using Satellite andAirborne Data ................................................................................................................ 145
Terrill W. Ray, Thomas G. Farr, Ronald G. Biota, and
Robert E. Crippen
The Red Edge in Arid Region Vegetation: 340-1060 nm Spectra ........................... 149
Tcrrill W. Ray, Bruce C. Murray, A. Chehbouni, and Eni Njoku
Temporal Changes in Endmember Abundances, Liquid Water and Water
Vapor Over Vegetation at Jasper Ridge ..................................................................... 153
Dar A. Roberts, Robert 0. Green, Donald E. Sabol, and
John B. Adams
Mapping and Monitoring Changes in Vegetation Communities of Jasper
Ridge, CA, Using Spectral Fractions Derived From AVIRIS Images .................... 157
Donald E. Sabol, Jr., Dar A. Roberts, John B. Adams,and Milton 0. Smith
The Effect of Signal Noise on the Remote Sensing of Foliar BiochemicalConcentration ................................................................................................................. 161
Geoffrey M. Smith attd Paul J. Curran
Application of MAC-Europe AVIRIS Data to the Analysis of Various
Alteration Stages in the Landdmannalaugar Hydrothermal Area (SouthIceland) ............................................................................................................................ 165
S. Sommer, G. LOrcher, and S. Endres
Estimation of Crown Closure From AVIRIS Data Using Regression
Analysis .......................................................................................................................... 169
K. Staenz, D. J. Williams, M. Truchon, and R. Fritz
Optimal Band Selection for Dimensionality Reduction of
Hyperspectral Imagery ................................................................................................. 173
Stephen D. Stearns, Bruce E. Wilson, and James R. Peterson
viii
Objective Determination of Image End-Members in Spectral
Mixture Analysis of AVIRIS Data ............................................................................... 177
Stefanie Tompkins, John F. Mustard, Carle M. Pieters,
and Donald W. Forsyth
Relationships Between Pigment Composition Variation and Reflectance
for Plant Species from a Coastal Savannah in California ........................................ 181
Susan L. Ustin, Eric W. Sanderson, Yaffa Grossman,
Quinn I. Hart, a_zd Robert S. Haxo
Atmospheric Correction of AVIR1S Data of Monterey Bay Contaminated
by Thin Cirrus Clouds .................................................................................................. 185
Jeannette vapz den Bosch, Curtiss O. Davis, Curtis D. Moblcy,
and W. Joseph Rhea
Investigations on the 1.7 _m Residual Absorption Feature in the
Vegetation Reflection Spectrum .................................................................................. 189
J. Verdebout, S. Jacquemoud, G. Andreoli, B. Hosgood,and A. Sieber
A Comparison of Spectral Mixture Analysis and NDVI for Ascertaining
Ecological Variables ...................................................................................................... 193Carol A. Wessman, C. Ann Bateson, Brian Curtiss, and Tracy L. Benning
Using AVIRIS for In-Flight Calibration of the Spectral Shifts of Spot-HRVand of AVHRR? ............................................................................................................. 197
Veronique Willart-Soufflet and Richard Santer
Laboratory Spectra of Field Samples as a Check on Two AtmosphericCorrection Methods ...................................................................................................... 201
Pung Xu and Ronald Greeley
Determination of Semi-Arid Landscape Endmembers and Seasonal Trends
Using Convex Geometry Spectral Unmixing Techniques ....................................... 205
Roberta H. Yuhas, Joseph W. Boardman, and Alexander F. H. Goetz
Slide Captions ................................................................................................................ 209
Volume 2: TIMS Workshop
Estimation of Spectral Emissivity in the Thermal Infrared ..................................... 1
David Kryskowski and J. R. Maxwell
The Difference Between Laboratory and in-situ Pixel-Averaged Emissivity:
The Effects on Temperature-Emissivity Separation ................................................. 5
Tsuneo Matsunaga
ix
Mineralogic Variability of the Kelso Dunes, Mojave Desert, California
Derived From Thermal Infrared Multispectral Scanner (TIMS) Data ................... 9
Michael S. RamsQf, Douglas A. Howard, Philip R. Christensen,and Nicholas Lancaster
Remote Identification of a Gravel Laden Pleistocene River Bed ............................ 13Dou_,las E. Scholen
Volume 3: AIRSAR Workshop
Interferometric Synthetic Aperture Radar Imagery of the
Gulf Stream .................................................................................................................... 1
T. L. Ainsworth, M. E. Cannella, R. W. Jansen, S. R. Chubb,
R. E. Carandc, E. W. Foley, R. M. Goldstein, and G. R. Valenzuela
MAC-91: Polarimetric SAR Results on Montespertoli Site ..................................... 5
S. Baronti, S. Luciani, S. Moretti, S. Paloscia, G. Schiavon,
and S. Sigismondi
AIRSAR Views of Aeolian Terrain ............................................................................. 9
Dan G. Blumberg and Ronald Greeley
Glaciological Studies in the Central Andes Using AIRSAR/TOPSAR ................. 13
Richard R. Forster, Andrew G. Klein, Troy A. Blodk, ett,and Bryan L. Isacks
Estimation of Biophysical Properties of Upland Sitka Spruce
(Picea Sitchensis) Pla nta tion_ ....................................................................................... 17Robert M. Green
Forest Investigations by Polarimetric AIRSAR Data in the Harz
Mountains ....................................................................................................................... 21
M. Keil, D. Poll, J. Raupenstrauch, T. Tares, and R. Winter
Comparison of Inversion Models Using AIRSAR Data for
Death Valley, California ............................................................................................... 25
Katht_tn S. Kierein-Youn_
Geologic Mapping Using Integrated AIRSAR, AVIRIS, and
TIMS Data ...................................................................................................................... 29Fred A. Krusc
Statistics of Multi-look AIRSAR Imagery: A Comparison of Theorv With
Measurements .......................................................................................... _..................... 33
J. S. Lee, K. W. Hoppel, and S. A. Mango
AIRSAR Deployment in Australia, September 1993:
Management and Objectives ....................................................................................... 37
A. K. Milne and I. J. Tapley
Current and Future Use of TOPSAR Digital Topographic Data for
Volcanological Research ............................................................................................... 41
Peter J. Mouginis--Mark, Scott K. Rowland, and H,Trold Garbeil
Preliminary Analysis of the Sensitivity of AIRSAR Images toSoil Moisture Variations ............................................................................................... 45
Rajah Pardipuranl, Williant L. Teng, James R. Wang, and
Edwin T. Engman
Unusual Radar Echoes From the Greenland Ice Sheet ............................................ 49
E. J. Rignot, J. I. van Zyl, S. J. Ostro, and K. C. Jezek
MAC Europe '91 Campaign: AIRSAR/AVIRIS Data Integration for
Agricultural Test Site Classification ........................................................................... 53
S. Sangiovanni, M. F. Buongiorno, M. Ferrarini, and A. Fiumara
Measurement of Ocean Wave Spectra Using PolarimetricAIRSAR Data ................................................................................................................. 57
D. L. Schuler
An Improved Algorithm for Retrieval of Snow Wetness UsingC-Band AIRSAR ............................................................................................................ 61
Jiancheng Shi, Jeff Dozier, and Helmut Rott
Polarimetric Radar Data Decomposition and Interpretation .................................. 65
Guoqing Sun and K. Jon Ranson
Synergy Between Optical and Microwave Remote Sensing to
Derive Soil and Vegetation Parameters From MAC Europe 91 Experiment. ....... 69
0. Taconet, M. Benallegue, A. Vidal, D. Vidal-Madjar, L. Prevot,
and M. Normand
SAR Terrain Classifier and Mapper of Biophysical Attributes .............................. 73
Fawwaz T. Ulaby, M. Craig Dobson, Leland Pierce, and
Kamal Sarabandi
Classification and Soil Moisture Determination of Agricultural Fields ................ 77
A. C. van den Broek and J. S. Greet
Relating P-Band AIRSAR Backscatter to Forest Stand Parameters ........................ 81
Yong Wang, John M. Melack, Frank W. Davis,Eric S. Kasischke, and Norman L. Christensen, Jr.
xi
Soil Moisture Retrieval in the Oberpfaffenhofen TestsiteUsing MACEurope AIRSAR Data ...................................................................................................85
Tobias Wever and ]ochen Henkel
Microwave Dielectric Properties of Boreal Forest Trees .......................................... 89
G. Xu, F. Ahem, and J. Brown
Slide Captions ................................................................................................................ 93
xii
N95- 23939
INTERFEROMETRIC SYNTHETIC APERTURE RADAR
IMAGERY OF THE GULF STREAM
T. L. Ainsworth 1'2, M. E. Cannella 1'3, R. W. Jansen 1, S. R. Chubb 1,
R. E. Carande 4, E. W. Foley 5, R. M. Goldstein 4 and G. R. Valenzuela 1
1 Naval Research Laboratory, Washington, D.C. 20375;
2 AlliedSignal Technical Services Corporation, Camp Springs, MD 20746;3 Syracuse University, Syracuse, NY 13244;
4 Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109;s Naval Surface Weapons Center, Bethesda, MD 20084
1. INTRODUCTION
Tile advent of interferometric synthetic aperture radar (INSAR) imagery
brought to the ocean remote sensing field techniques used in radio astronomy. Whilstdetails of the interferometry differ between the two fields, the basic idea is the same:
Use the phase information arising from positional differences of the radar receivers
and/or transmitters to probe remote structures. The success of airborne INSAR
methods (Goldstein et al., 1987) provided ample incentive to investigate numerousother applications, e.g. topographic mapping (Zebker et al., 1986), surface ocean
currents (Goldstein et al., 1989) and internal waves (Thompson et al., 1993). In thispaper, we apply for the first time INSAR methods to the Gulf Stream boundary. A
primary advantage of the INSAR technique when applied to ocean surfaces is the
ability to observe the motion of surface scatterers.
The interferometric image is formed from two complex synthetic aperture
radar (SAR) images. These two images are of the same area but separated in time.
Typically the time between these images is very short -- approximately 50 msec forthe L-band AIRSAR. During this short period the radar scatterers on the ocean
surface do not have time to significantly decorrelate. Hence the two SAR images will
have the same amplitude, since both obtain the radar backscatter from essentially the
same object. Although the ocean surface structure does not significantly decorrelatein 50 msec., surface features do have time to move. It is precisely the translation of
scattering features across the ocean surface which gives rise to phase differences
between the two SAR images. This phase difference is directly proportional to the
range velocity of surface scatterers. The constant of proportionality is dependent
upon the interferometric mode of operation. In our case, the total phase differencebetween the two SAR images is
27rt
A_ _ Vair_ Urang e
where g is the spacing between the radar receivers, ,k is the radar wavelength, vair is
the aircraft velocity and ur_,_ is the component of the scatterer's velocity in the
range direction. The motion of the scatterers may arise from ocean currents, internal
wave motions, winds or most generally all of the above. Identifying these different
componentsof ur_nge, without recourse to additional information, is a formidabletask.
One immediately sees several possible limitations to the INSAR technique.The time between images must be short enough that the ocean surface does notdecorrelate, otherwise the phase difference contains no new information. Also the
time between images must be long enough so that the surface features have time to
move and thus provide a phase difference. We implicitly require that the SAR imageshave been corrected for the aircraft sidewards drift and yaw, both of which will
produce (unwanted) phase differences. Therefore as a practical matter the phase
difference arising from the surface motion should be greater than the errors in
compensating for aircraft motion -- the signal-to-noise ratio should be large. Perhaps
more irksome than the above is the problem of large surface velocities, when the
phase difference is greater than 27r. The interferometric image determines the phase
differences modulo 27r, therefore many velocities map onto the same phase difference.
Typically, range velocities of scatterers of approximately 0.0 m/sec., 2.7 m/sec.,
5.4 m/see., etc. all yield a zero phase difference in the interferometric image. One
possible means of lifting this phase ambiguity is to use multi-frequency and/ormulti-baseline interferometry. These methods have been discussed by Carande, 1992
and Carande el el., 1991. Therefore the interpretation of INSAR images may require
other information, additional assumptions and/or modeling.
2. GULF STREAM IMAGES
The particular interferometric images which we have analyzed are from the
1990 AIRSAR flight of July 20 th, previously identified as G-Stream NI 120-1
(Kobrick, 1990; Valenzuela et al., 1991). The complex interferometric image is the
product of one complex SAR image with the complex conjugate of the other.
Therefore the amplitude of the INSAR image is the product of the individual SAR
image amplitudes. The phase of the INSAR image is the phase difference (modulo
27r) between the two SAR images. The amplitude and phase of the INSAR image areshown in Figure 1. The amplitude image shows no detailed structure and provides
only hints of large features. In contrast the phase image clearly depicts the Gulf
Stream boundary. The Gulf Stream boundary is the one large feature that may be
seen in the amplitude image. In addition, the phase image clearly shows the ResearchVessel Cape Henlopen, her wake and other sxnaller surface structures. We have
interpreted the dark-light banding as internal wave motion. The orientation of these
images is independently verified by the ship's wake and log--- she was southbound at10 knots when the INSAR image was taken.
This comparison of INSAR phase and amplitude images may downplay toomuch the value of complex SAR imagery. Milman et el., 1990 have developed and
applied ambiguity function techniques to complex SAR images to obtain information
about surface velocities. However, the amplitude of SAR images conveys no relevant
information about velocities (other than velocity bunching effects) and, as seen inFigure 1, little information about small-scale surface features.
The complementary nature of SAR and INSAR images produces a powerfuloceanographic remote sensing tool. INSAR is sensitive to surface velocities whereas
SAR amplitudes depend strongly on the surface shape. Presumably these twofeatures (shape and velocity) are, in at least some cases, correlated. Therefore
employing the additional information which interferometry provides will assist theinterpretation of the region surveyed.
wa
fl_ _ ._ * _'_ _4_"_ I
Figure 1: The INSAR amplitude (left) and phase (right) images for run G-Stream NI
120-1 taken July 20, 1990 are displayed. The area of each image is approximately 5 km
in azimuth and 10 km in range. See text for additional discussion.
3. GROUND TRUTH INFORMATION AND MODELING
Ground truth information is available in the form of buoy measurements of
the wave spectrum and direction. Two buoys were deployed both before and after the
INSAR image of Figure 1 was taken. We have Fourier analyzed the phase image to
determine the spectrum of the wave-like structures seen throughout the image. The
frequency and direction are directly compared to the buoy data. The Henlopen's
wake also may aid in our interpretation of surface features. Thus the available ground
truth in conjunction with the INSAR image provides useful constraints for modeling
Gulf Stream boundary features.
We are currently modeling both INSAR and SAR radar return from ocean
surfaces. This is an ongoing project and results will be reported at a future date.
4. CONCLUSIONS
Our initial investigation has yielded several interesting characteristics of
INSAR imagery: The wave spectrum is clearly seen in tile phase image but not in the
amplitude. The boundary of the Gulf Stream is a strong linear feature -- barely
resolved in tile amplitude image -- which dominates the phase image. Similarly, the
Henlopen's wake is easily seen in the phase image. Of course, the ship is clearly
resolved in both tile amplitude and phase images.
ACKN OWLEDGEMENTS
We thank J. S. Lee aud D. Schuler from NRL for many helpful insights and
informative discussions. Wq, also thank S. Madseu from J PL for performing the initial
image processing.
REFERENCES
(_arande, R. E., R. M. Goldstein, Y. Lou, T. Miller and K. Wheeler, 1991,
"Dual-frequency Along-track lnterferometry: A Status Report," Proceedings ofthe Third Airborne Synthetic Aperture Radar Workshop, JPL Publication 91-30,
Jet l)ropulsion Laboratory, Pasadena, California, pp. 109 116.
Carande, R. E., 1992, "Dual Baseline and Frequency Along-track
luterferometry," Proceedings of lGARSS '92, pp. 1585 1588.
Goldstein, R. M. and 1t. A. Zebker, 1987, "Interferometric Radar Measurements
of Ocean Surface Currents," Nature, 328, pp. 707 709.
Goldsteiu, R. M., T. P. Barnett and H. A. Zebker, 1989, "flemote Sensing of
Ocean Currents," Science, 246, pp. 1282-1285.
Kobrick, M., 1990, "AIRSAR Data Digest Summer 1990," JPL D-8123
(internal document), Jet Propulsion Laboratory, Pasadena, CA.
Mihnan, A. S., A. O. Scheftter and J. B. Bennett, 1990, .1. Geophys. Res., 98,pp. 911- 925.
Thoml)SOn, D. R. and J. R. Jeusen, 1993, "Synthetic Aperture Radar
Interferometry Applied to Ship-generated Internal Waves in the 1989 Loch
Linuhe Experiment," J. Geophys. Res., 98, pp. 10259-10269.
Valenzueta, G. R., R. P. Mied, A. R. Ochadlick, M. Kobrick, P. M. Smith, F.
Askari, R. J. Lai, D. Sheres, J. M. Morrison and R. C. Beal, 1991, "The July
1990 Gulf Stream Experiment," PTvceedings of IGARSS '91, pp. 119-122.
Zebker, 1t. A. and R. M. Goldstein, 1986, "Topographic Mapping from
Interferometric Synthetic Aperture Radar Observations," J. Geophys. Res., 91,pp. 4993-4999.
4
MAC-91: POLARIMETRIC SAR RESULTS ON MONTESPERTOLI SITE N95- 23940
S. Baronti (*), S . Luciani (**), S. Moretti (***),S. Paloscia (*), G. Schiavon (**) and S. Sigismondi (***)
(*) lstituto di Ricerca sulle Onde Elettromagnetiche - CNR
Via Panciatichi 64, 50127 Firenze (I)
(**) Dipartimento di lngegneria Eiettronica, Universita Tor Vergata
Via della Ricerca Scientifica, 00133 Roma (I)
(***) Dipartimento di Scienze della Terra, Universit_t di FirenzeVia La Pira 4, 50121 Firenze (I)
1. INTRODUCTION
The polarimetric Synthetic Aperture Radar (SAR) is a powerful sensor for high
resolution ocean and land mapping and particularly for monitoring hydrological
parameters in large watersheds. There is currently much research in progress to
assess the SAR operational capability as well as to estimate the accuracy achievable
in the measurements of geophysical parameters with the presently available
airborne and spaceborne sensors. An important goal of this research is to improve
our understanding of the basic mechanisms that control the interaction of electro-
magnetic waves with soil and vegetation. This can be done both by developing
electromagnetic models and by analyzing statistical relations between backscatte-
ring and ground truth data. A systematic investigation, which aims at a betterunderstanding of the information obtainable from the multi-frequency polari-
metric SAR to be used in agro-hydrology, is in progress by our groups within the
framework of SIR-C/X-SAR Project and has achieved a most significant
milestone with the NASA/JPL Aircraft Campaign named MAC-91. Indeed this
experiment allowed us to collect a large and meaningful data set including multi-
temporal multi-frequency polarimetric SAR measurements and ground truth.
This paper presents some significant results obtained over an agricultural flat area
within the Montespertoli site, where intensive ground measurements were carried
out. The results are critically discussed with special regard to the information
associated with polarimetric data.
2. EXPERIMENT DESCRIPTION
During the Multisensor Airborne Campaign (MAC-91), the site of Montespertoli(Italy) was imaged on three different dates (22 June, 29 June and 14 July) by the
airborne JPL polarimetric SAR (AIRSAR) operating at P-, L- and C- band
(Canuti et al., 1992). Three passes were performed during each flight data, in
order to cover the most significant sub-areas with different incidence angles 0 be-
tween 20* and 50*. Nominal pixel sizes obtained on the 16 looks images were 6.6
m in range and 12 m in azimuth. The test site covers the basins of two small rivers
and contains a relatively large flat area, along the border of the Pesa river, which
includes agricultural fields of various crops (i.e. alfalfa, colza, corn, sorghum, sun-
flower, wheat) bare soils, olive-yard, vineyards and a few small forests. The area
was equipped with two trihedral corner reflectors of 180 cm and one of 240 cm,supplied by the JPL, and deployed along the Pesa river. In addition, a few
"extended homogeneous targets", namely three plots of bare soil with different
surface roughness, one field of alfalfa and one of sunflower, had been specifically
prepared, in the same sub area, to act as "distributed" calibrators. The ground data,which were collected on selected fields at the same time as the remote sensing
measurements, regarded all the significant vegetation and soil parameters, such as
leaf area index (LAI), plant water content (PWC), dimensions of leaves and stalks,
soil moisture content and roughness. During the experiment the average
gravimetric soil moisture of the first 5 cm layer changed from about 15-20 % (on
June 22) to about 10% (on July 14). At the same time the leaf area index (m2/m 2)of sorghum and corn increased from - 0.5 to - 3.5, that of sunflower from ~ 0.5to ~ 3.5. Alfalfa leaf area index ranged from ~ 0 to - 4. Wheat and colza were inthe ripening stage and therefore their LAI was almost 0; some of these fields wereharvested at the time of the last flight.
3. DATA ANALYSIS
Calibrated (amplitude and phase) polarimetric data, collected over the agriculturalarea on alfalfa, corn, colza, sunflower, sorghum, bare soil, olives, vineyard and ona few forested areas, at incidence angles of 35 ° and 50 °, have been analyzed.
Background - Scattering sources may be grouped in broad categories, eachassociated to typical scattering behaviors, as shown in recent works (e.g. Ulaby etal. 1990, Van Zyl 1989). In particular: a) Soil response is a surface scatteringeffect. This effect is important for bare soils and, in general, at low frequencies,where many agricultural crops are rather transparent, ooRn is appreciably higher
than O*aR and oOnv is low. At lower frequencies (P and L band) oOvv> OOHH,as predicted by the Small Perturbation surface scattering model (Ulaby et al.
1982), while at higher frequencies (C band) O*vv = O*nH. b) Vertical structures,like forest trunks and crop stalks, produce double-bounce scattering. In general,this mechanism is important at P band for forests and at L band for some
agricultural crops like corn and sunflower. O*uv is low, as in the soil scattering
case, but, differently from the soil, o*nn is generally higher than O*vv and thelarge difference between o*RL and o*aR disappears, c) Ensembles of inclinedcylindrical structures, like forest branches and crop stems, produce volumescattering with an appreciable presence of multiple scattering. The differenceso*nH - O*nv and O*vv - O*nv are much lower than those for soil and vertical
structures. In circular polarization, O*RL = o*aR. d) Ensembles of inclined planarstructures, like leaves, also produce volume scattering; however, an appreciable
amount of single "facet" scattering is present, o*rm - O*Hv and O*vv - O*nvdifferences are relatively low, similarly to the cylinder case, but at circular
polarization, the "facet" effect generates appreciable positive O*rtt, - O*Rrtdifferences. The scatterer dimensions have important effects too. Let branches andstems be represented as ensembles of cylinders with the length proportional to theradius. For a canopy of equal cylinders, the backscatter coefficient changes withthe cylinder length (expressed in wavelengths) and shows a maximum (Karam etal. 1992). It follows that, for each band, there is a range of cylinder dimensionsgenerating a dominant contribution to the backscatter. For leaves, which can bedescribed as discs, scattering is dependent on thickness and moisture content(Ferrazzoli et al. 1993).
4. SURFACE TYPE DISCRIMINATION
Some significant backscattering features of agricultural surfaces can be inferred
from Fig. 1 which represents the cross-polarized backscattering coefficient (O*nv)of well developed (Plant Water Content - PWC > 1 Kg/m 2) crops measured at L-
band and 0 = 35*, as a function of the corresponding o*rl v at C-band. At L-band there is a continuous increase of backscatter as the dimensions of scatteringelements increase from the 'small leaf' crops such as alfalfa (A) and sorghum (S)to the bigger olive-yards (O) and forest trees (T), passing through 'broad leaf'crops such as sunflower (F) and colza (R). At C-band, bare soil can be discrimi-nated from crops among which colza, which has a well ramified structure, has thehighest backscattering value.The relatively strong capability of C-band backsc-attering to discriminate agricultural surfaces is improved if circular polarizationdata are used as well (Baronti et al. 1993). Indeed, bare soils and plants character-ized by relatively large leaves, like sunflower, corn, sorghum, show a "facet" sca-
ttering with OORL > O'*RR. while, in the other fields, OORL _ OORR . However
colza crops show rather high backscatter
at both RR and RL polarizations, due to
their scattering elements (stems, pods)which are numerous and relatively large,
while the elements of wheat and alfalfa
crops produce lower backscattering. Itmust be noted that, unlike L-band,
discrimination is feasible also in the case
of moderate growth. Figure 2 shows
o*HV versus o'Ha measured at P-bandand 0 = 35*. According to the previous
considerations the scattering in HV pol.
is mainly due to inclined and relatively
large cylindrical elements, like branchesof forest, oliveyard and vineyard, while
the agricultural vegetation is rather
transparent. Three classes of o*HV val-
ues may be identified corresponding to:
forests (T), olive-yard and vine-yard (O
and V) and agricultural fields (A). Ac-
cordingly P-band backscattering, whichis
-lO
aurw 22. 1991
-15 F_
iSvv F
"_-2o VIOBO0 R
o / vV/m _, will F
-25-
-_-:,2 ' -_ ' -_a ' -_e ' -_4 ' -_2 ' -_o ' -8C Band (HV)
Figure 1 - Backscattering Coefficient
(u*) at L-band vs.o* at C-band (HV
pol. 0=35*) Labels: B=bare soil, T=-forest, V=vine-yard, O=olive-yard,
W=wheat, A= alfalfa, S=sorghum,
F=sunflower, R=colza, I=uncropped
-1@
> -1_,I
0
-20-
Erl
-25-
_T YT
V
VV V
vY VoY.vv _,Voo
_V v VA_;AAA. A A
-_s ,_ -'50 :'_s -'1oSigmaO HH (dB)
-5
Figure 2 - a*nv vs. aoan at P-band,0=35"
dominated by the tree branches and thesoil trunk interaction, appears rather
effective in discriminating broad classes
of surfaces (forest, olive-yard, vine-yard,
agricultural fields) but their utility indiscriminating the various agricultural
crops is poor. At L-band, where scat-
tering is mainly due to stems and large
leaves, the SAR data appear more suitable
for separating the 'broad ieaff crops (corn
and colza) from other crops; this discrim-ination is improved using C-band data
which are particularly sensitive to small
stems and leaves (Coppo et al. 1989). Note
that a discrimination between alfalfa and
wheat fields does not appear feasible at
C-band; however, recent works indicate
that it becomes possible by means of both
active and passive remote sensing systems
if higher frequencies (X- and K- bands) are used (Ferrazzoli et al. 1992).
SENSITIVITY TO SOIL MOISTURE AND VEGETATION BIOMASS
Soil Moisture - The measurement of soil moisture (SMC) with radar has been the
subject of much research over the past years. However, a reliable extraction ofsuch information from SAR data is still questionable in that the signal is strongly
influenced by surface roughness and vegetation cover. In our case data collected
at L and C-band show very little correlation with SMC even at 0 = 35*. A slightly
better result has been obtained at P-band where the o* I_It has shown a sensitivity
to SMC equal to 1.0 dB/% SMC and a correlation coefficient equal to 0.7.
Herbaceous crops - In this experiment the agricultural vegetation was rather
transparent at P-band while, at C-band, it generated considerable scatteringeffects even in the case of moderate growth. Indeed, the highest correlations with
plant growth have been noted at L-band. On the other hand o*t_ v, which is low
for bare soils and is mainly associated tothe presence of inclined stems and large -16leaves, is a relatively good growth indi-
cator. Figure 3 shows L-band OOHv asa function of the Plant Water Content _-20
(PWC, kg/m2). Samples of sorghum (S),corn (C), sunflower (F) and bare soil >
-24have been included. Very rough (ploug- o
ohed) bare soils are labeled 'PB', while E
moderately rough (tilled or rolled) bare _-2asoils are labeled "TB'. The correlation is
generally good; the lowest o°HV values
are associated with moderately rough -32bare soils and with sorghum fields at avery early stage of growth, while thehighest values are associated with welldeveloped sunflower crops.Ploughedbare soils show relatively high val-ues,indicating that the roughness plays a
rr
PB CF
$S
f csS
FF
6 i _ 3 4PWC
Figure 3- o*lav vs. PWC (Kg/m z) atL-band, 0=50*
role similar to that of vegetation growth. The dynamic range is notable, in the
order of 10 dB. The plant growth may be also related to O*HH - O*VV at L-band.In fact, for bare soils, O*HH < O°VV while, for developed crops, the soil plantdouble bounce scattering, which may be rather important at L-band, produces apositive O*HH - O*VV difference.
CONCLUSIONS
Polarimetric SAR measurements carried out over forests, oliveyards, vineyards andagricultural fields indicate that P-band data are effective in discriminating amongbroad surface categories, L-band data allow identification of well developed sun-flower, corn and colza crops, and C-band data discriminate among different kinds
of herbaceous crops even in the case of moderate growth. A fairly good sensitivityof o°HV and O*Hn - O*VV tO the Plant Water Content of some agricultural speciesis observed at L-band.
ACKNOWLEDGMENT
This work was partially supported by the Italian Space Agency
REFERENCES
Baronti, S. F. Del Frate, P. Ferrazzoli and S. Paioscia, "Interpretation of Polarimetric MAC-91 data
over Montespertoli Agricultural area," Proc of the 25th Intern. Symposium, Remote Sensing and Global
Change, Gral (Austria), 1993.
Canuti, P., G. D'Auria, P. Pampaloni, and D. Solimini "MAC-91 on Montespertoli: an experiment for
agro-hydrology", Proc. of the IGARSS Symposium, Houston, Texas, 1992, 1744-1746.
Coppo, P., P.Ferrazzoli, G.Luzi, P.Pampaloni, G.Schiavon, D.Solimini 1989, "Microwave remote
sensing of land surfaces with airborne active and passive sensors," Proc. of the IGARSS Syrup.,
Vancouver (Canada), 1989, 806-809.
Ferrazzoli, P.,D. Solimini, G. Lugi, and S. Paloscia, "Model analysis of backscatter and emission from
vegetated terrains," J. of Electromagnetic Waves and Applications, vol. 5, 175-193, 1991
Ferragzoli, P., L. Guerriero, S. Paloscia, and P. Pampaloni, "Modeling X and Ka band emission from
leafy vegetation," Proc. of the URSI Microwave Signature Conf., Igls-lnnsbruk, 1992, 2B14-2B17.
Karam, M.A., A.K. Fung, R.H. Lang, and N.S. Chauhan, "A microwave scattering model for layered
vegetation," IEEE Trans. Geosci. Remote Sensing, vol. GE-30, 767-784, 1992.
Uiaby, F.T., R.K Moore, and A.K. Fung, Microwave Remote Sensing: Active and Passive. Vol.II -
Surface Scattering and Emission Theory, Addison-Wesley, Reading, Massachusetts, 1982.
Ulaby, F.T., and C. Elachi, editors, Radar polarimetry for geoscience applications, Norwood, MA:Artech House, 1990
Van Zyl, J.J., "Unsupervised classification of scattering behavior using radar polarimetry data", IEEE
Transactions on Geoscience and Remote Sensing , vol. GE-27, pp. 36-45. 1989.
N95- 23941
AIRSAR VIEWS OF AEOLIAN TERRAIN
Dan G. Blumberg and Ronald Greeley
Department of GeologyArizona State UniversityTempe Az. 85287-1404
1. OVERVIEW
Remote sensing observations of terrestrial dune morphologies provideinformation on sand transport, climate, and soil processes associated with aeolianprocesses. In this study AIRSAR images of the Stovepipe Wells Dune Field, California,were examined to assess the degree to which dune types can be determined from radar
images and the factors leading to the formation of four different dune types in onelocality. The images were acquired during the Geologic Remote Sensing Field
Experiment (GRSFE; Evans et al., in press) and later. These images provide a uniqueopportunity to enhance the knowledge of how radar images allow discrimination of duneforms.
2. RADAR DATA
Radar images were obtained in September 1989 and May 1992 by the JetPropulsion Laboratory synthetic aperture radar - AIRSAR. The images have a resolutionof -12 m per pixel and are acquired in C-band (_, = 5.6 cm), L-band (L = 24 cm), andP-band (_ = 67 cm). All images were calibrated using POLCAL (van Zyl et al., 1992).Although the entire Stokes matrix is saved for the images and any polarization can beextracted, only the co- and cross-polarized data were extracted (HH; VV; HV-VH).Figure 1 (See Slide 12) shows an image of the total power return at Stovepipe Wellsfor C, L, and P-band in HH polarization.
The illumination was from the west and the flight direction from north to south.
The look angle was 45 °.
3. OBSERVATIONS
The Stovepipe Wells Dune field covers an area of- 100 km 2 in northern Death
Valley, California (117°06 W 36 ° 39 N). Four dune types, each characteristic of adifferent wind regime, occur at Stovepipe Wells: reverse dunes, star dunes, transverse, andlinear dunes. In addition to dunes, C-band images reveal a mantle of sand around the dunefield. Field observations show that local saline crusts in the sand mantle cause microscale
roughness variations that give a mottled radar backscatter cross section that is prominentin C-band but less pronounced in P and L-bands.
3.1 Reverse dunes
The reverse dunes are low (typically < 5 m high), formed by temporal reversalsof the wind direction through the valley (N - S). The reverse dunes are observed in theradar images as dark and bright linear features on the northeast part of the field. The dunes
appear as dark narrow streaks aligned east-west with bright interdune regions. The lowreverse dunes are not seen in P-band, but are visible in C-band. They are best observed in
the H-H images, and not seen in the cross-polarized image.
3.2 Star dunes
The star dunes have a maximum height of 40 m above the surrounding surlhce
and arc lormed by polymodal winds transporting sand to the center of the dune field. Stardunes have at least three flanks which create a trihedral corner reflector effect in radar
images. The result is an extremely bright speckle from the interdune zone facing the
antenna. These dunes are easily recognized in AIRSAR images of Stovepipe Wells
because of the quasi-speckular return from the corner between the dune flanks. The quasi-
speckular signature is visible in all wavelengths.
3.3 Transverse Dunes
Transverse dunes are lound in the northern part of the dune field. These are
fornmd by unimodal winds blowing from the north. The winds from the south (that are
partly responsible for the reverse dunes) are blocked by a mountain and by the large star
dunes. The transverse dunes appear in the radar images as alternating bright and dark
stripes. These dunes exhibit bright lee slopes and dark narrow sloss slopes. Transverse
dunes tire visible in all wavelengths but are not seen in the cross-polarized data.
3.4 Linear dunes
In the west part of the dune field there is a set of linear dunes which me formed
by hi-directional winds coming from the northeast and southeast. Biota and Elachi (1981)
state that tim interdune spacing needs to bc at least 3 to 4 picture elements wide to be able
to detect linear dunes. The interdunal spacing of these dunes is greater than 4 pixcls. In
the 1992 image these dunes appear as narrow reflective streaks. Because the linear dunes
arc in the very near range, their form is more difficult to resolve in this image than other
dune lypes.
3.5 Vegetated dunes
Very sparse vegetation exists at the Stovepipe Wells Dune Field. The
vegetation is primarily in clusters over the sand mantle and in an area where tim dune
torms change from reverse to star dunes. While vegetation can be seen in all images it is
most pronounced in the cross-polarized images. Comparison of the vegetation in the
radar images to aerial photographs from 1948 do not show any spatial change in the
vegetation pattern.
4. Conclusion
The Stovepipe Wells Dune Field provides a unique opporttmity to observeseveral dune forms in one scene. These forms include reverse dunes, star dunes,
transverse, anti linear dunes. A sand mantle surrounds the dune field and can also be
observed in the radar image. Dune types were discriminated best in co-polarized channels.
Three major wind directions are responsible for the various dune forms. Wind
from the north and south are responsible for the reverse dunes, winds from north the for
the Iiansverse dunes, and the north - south and westerly winds form the slar dtlnes. The
\rinds also reflect the topographic configurathm of this part of tile valley.
Vegetation over the dunes was most pronot, nced in the cross-polarized ilna_es.
I_ancastcr ct al. (1992) finds that cross-polarized images arc most useful in differcntiatin::
act+re from inactive dunes. This is because the vegetation backscatter signature is
prcscnl over inactive dunes.
Future studies shouhl include multiple look and incidence angles t_ dote, mine it
the dune forms can still be seen at other angles.
10
Figure 1. AIRSAR image of Stovepipe Wells Dune Field, Death Valley, California. The image showstotal power for C, L, and P-bands
1!
References cited:
Biota R. and C. Elachi, 1987, "Spaceborne and Airborne Imaging Radar Observations of
Sand Dunes", J. Geophys. Res., 86 (B4), 3061-3073.
Evans D.L., T.G. Farr, M. Vogt, C. Bruegge, J. Conel, R.E. Arvidson, S. Petroy,J.J. Plaut, M. Dale-Bannister, E. Guiness, R. Greeley, N. Lancaster, L. Gaddis,J. Garvin, D. Deering, J. R. Irons, F. Kruse, D. J. Harding, submitted, "The GeologicRemote Sensing Field Experiment", IEEE Transaction of Geoscience and RemoteSensing.
Lancaster N., L. Gaddis, and R. Greeley, 1992, "New Airborne Imaging RadarObservations of Sand Dunes: Kelso Dunes", California, Remote Sens. Environ., 39,
pp. 233-238.
van Zyl J.J., C.F. Burnette, H.A. Zcbker, A. Freeman, J. Holt, 1992, Polcal Users'Manual, JPL D-7715 version. 4.0., pp. 48.
12
N95- 23942
GLACIOLOGICAL STUDIES IN THE CENTRALANDES USING AIRSAR/TOPSAR
Richard R. Forster, Andrew G. Klein, Troy A. Blodgett
and Bryan L. IsacksCornell University
Department of Geological SciencesSnee Hall
Ithaca, New York 14853-1504
1. INTRODUCTION
The interaction of climate and topography in mountainous regions is
dramatically expressed in the spatial distribution of glaciers and snowcover.Monitoring existing alpine glaciers and snow extent p.rovides insight into thepresent mountain climate system and how it is changing, while mapping thepositions of former glaciers as recorded in landforms such as cirques andmoraines provide a record of the large past climate change associated withthe last glacial maximum. The Andes are an ideal mountain range in whichto study the response of snow and ice to past and present climate change.Their expansive latitudinal extent offers the opportunity to study glaciers indiverse climatic settings from the tropical glaciers of Peru and Bolivia to theice caps and tide-water glaciers of sub-polar Patagonia.
SAR has advantages over traditional passive remote sensinginstruments for monitoring present snow and ice and differentiating morainerelative ages. The cloud penetrating ability of SAR is indispensable for
perennially cloud covered mountains. Snow and ice facies can bedistinguished from SAR's response to surface roughness, liquid watercontent and grain size distribution. The combination of SAR with acoregestered high-resolution DEM (TOPSAR) provides a promising tool formeasuring glacier change in three dimensions, thus allowing ice volumechange to be measured directly. The change in moraine surface roughnessover time enables SAR to differentiate older from younger moraines.
Polarimetric SAR data have been used to distinguish snow and ice
facies (Shi et al., 1991) and relatively date moraines (Forster et al., 1992).However, both algorithms are still experimental and require ground truthverification. We plan to extend the SAR classification of snow and icefacies and moraine age beyond the ground truth sites to throughout theCordillera Real to provide a regional view of past and present snow and ice.The high resolution DEM will enhance the SAR moraine dating techniqueby discriminating relative ages based on moraine slope degradation (Bursik,1991).
2. 1993 FIELD CAMPAIGN
The 1993 South American AIRSAR campaign acquired data at foursites in the Peruvian and Bolivian Andes. TOPSAR data was acquired at all
13
sites. Threeof the four sites, the Quelccaya Ice Cap, the Cordillera Blanca,both in Peru, and the Cordillera Real, Bolivia were targeted for theirnumerous modern glaciers and Pleistocene glacial landforms. The fourth,Potosi, Bolivia was chosen for its well preserved multiple moraines andalluvial outwash plains. The Cordillera Real was chosen as the groundtruth site because of the easy access to its glaciers.
Ground truth data were acquired on two glaciers in the CordilleraReal during the AIRSAR flight. The two glaciers reside on adjacentmountains of the Cordillera north-east of La Paz. We recorded the
following data from the Chacaltaya glacier: surface roughness profiles,snow depth transacts, and snow pit profiles of density, temperature,grainsize distribution, and wetness. Measurements were taken on two daysprevious to the flight, the day of the flight and the day after the flight.Metrological data were recorded on the flight day and the following day.French collaborators from ORSTOM took a set of similar measurements on
their research glacier, the Zongo glacier, during the AIRSAR flight. Fourcomer reflectors (manufactured in Bolivia) were deployed and theirpositions determined with GPS on and around the Chacaltaya glacier for co-regestation. The entire Cordillera was imaged with six swaths, five parallelto the mountain range and one oblique, optimizing the viewing geometry ofthe two neighboring ground truth glaciers.
During the two weeks after the SAR flight we collected samplesfrom original surfaces of moraine boulders at several locations in the
Cordillera Real and at the Potosi site for 36C1 cosmogenic ray dating. Thiswill provide numerical dates for the SAR derived moraine chronology.
3. ANTICIPATED RESULTS
The measured surface roughness and dielectric properties (snowwetness, grain size distribution and density) of the snow and ice will beused to interpret the polarization signatures obtained from different snowfacies present on the glaciers. We will also map the snow facies boundarieson the test glaciers and extrapolate these results to glaciers throughout theCordillera. We anticipate this research will provide the ground work forestablishing a three dimensional base line for future monitoring of annualsnowline and glacier positions over a regional setting. This DEM will beused for comparison with our present digitized topography and SPOTderived DEMs. The high resolution DEM derived from TOPSAR combined
with a regional moraine chronology will allow detailed glacierreconstruction and ice volume calculations for known times during thePleistocene. The AIRSAR coverage of the Cordillera Real will overlap witha SIR-C/X-SAR site providing an instrument and temporal comparison.Data from the Patagonian SIR-C/X-SAR supersite along with a similarground campaign will allow a direct comparison of polarimetric SARresponse to high elevation tropical glaciers and mid-latitude low elevationglaciers.
14
4. REFERENCES
Bursik,M., 1991,Relativedatingof morainesbasedon landformdegradation,LeeVining Canyon,California.Quaternary Research,35, pp. 451-455.
Forster, R. R., Fox, A. N., Isacks, B. I., 1992, Preliminary results ofpolarization signatures for glacial moraines in the Mono Basin,Eastern Sierra Nevada, Summaries of The Third Annual JPLAirborne Geosciences Workshop, JPL Publication 92-14 vol. 3, pp.40-42.
Shi, J., Dozier, J., Rott, H., Davis, R. E., 1991, Snow and glacier
mapping in alpine regions with polarimetric SAR. ProceedingsIGARSS' 91, IV, pp.2311-3214.
15
N95- 23943
ESTIMATION OF BIOPtlYSICAL PROPERTIES OF UPLAND SITKA
SPRUCE (PICEA SITCHENSIS) PLANTATIONS.
Robert M. Green
Department of Geography, University College of Swansea,
Singleton Park, Swansea, SA2 8PP, U.K.
1. INTRODUCTION.
It is widely accepted that estimates of forest above-ground biomass are required
as inputs to forest ecosystem models (Kasischke and Christensen, 1990), and that SAR
data have the potential to provide such information (e.g. Kasischke et al., 1991; LeToan
et al., 1992). This study describes relationships between polarimetric radar backscatter and
key biophysical properties of a coniferous plantation in upland central Wales, U.K. Over
the test site topography was relatively complex and was expected to influence the amountof radar backscatter.
) TEST SITE AND DATA
As part of the NASA MAC Europe (Curran and Plummer, 1992) the JPL
AIRSAR was flown over the Tywi forest, in central Wales, U.K. This is an area of
coniferous forest plantations, consisting of Sitka spruce (Picea sitchensis), Lodgepole pine
(Pinus contorta var. latofoliar) and Japanese larch (Larix kaempfei), surrounding the Llyn
Brianne reservoir. It is an upland region characterized by variable topography of up to500m above mean sea level with some valley slopes of 45 ° or more. For this study only
areas of Sitka spruce were analyzed since this was the dominant species with the widest
range of planting dates on a variety of slope/aspect orientations relative to the SAR. All
the species are densely stocked with planting spacing typically 2 metres. The ground data
comprised results of an intensive field survey for a limited number of stands; and for a
much larger area, the planting date of forest compartments was known from stocking
maps.
The AIRSAR images consisted of a slant range pixel resolution of 3.4m and
incidence angle ranged from approximately 20 ° near-range to between 55 ° and 60 ° far-
range. The Stokes matrix data were analyzed using POLTOOL, a polarimetric SAR data
processing package. P-band data were subject to interference at all polarizationcombinations and were consequently not used in the analysis. This is a disappointingsituation since some authors have found P-band data to be most sensitive to forest biomass
(e.g. LeToan et al., 1992).
2. BACKSCATrER RELATED TO STEM BIOMASS ESTIMATES
Estimates of key biophysical parameters were derived for forest stands within the
Cefn Fannog plantation managed by the U.K. Forestry Commission. Data available from
the Forestry Commission included surveys performed in 0.01ha plots at regular intervalsin the area. From standard forestry mensuration data the average basal area, stem volume
and stem biomass for each stand were derived. The average intensity of backseatter
(amplitude) for a group of pixels corresponding to a forest stand was extracted from theAIRSAR imagery, synthesized into HH, HV and VV polarizations using POLTOOL.
t,n mcm B.at nua:n17
0.8 / _ CHH
® CHV
0.6 0- cvv
_ 0.4
(a)
r - 0.39
r-0.52
r - 0.47
0.2
0 i i
8O 120
Stem biomass (tons/ha)
(b)
i
160 2OO
0.8-
i
o 40
LHH r - 0.48
0"" [.J-fV r - 0.54 • •
o ! i i ,
0 40 80 120 160 200
Stem biomass (tons/ha)
0.6-
" 0.4-
0.2-
Figure 1. Plots of amplitude versus average stand biomass using data
synthesized into HH, ltV, and W polarizations in (a) C-band, and (b) L-band.
Figure 1 shows plots of C-, and L-band backscatter versus stem biomass. The
amount of C-band backscatter was lower than L-band. Luckmann and Baker (1993), usingthe same data set, attribute this to a lack of scatterers at the same size as C-band
wavelengths. In Figure la the best-fit lines do not show any significant increasingrelationships in any of the polarization combinations. This is to be expected if it is
assumed that C-band scattenng takes place in the upper levels of the canopy, since all the
stands which were sampled had uniform canopy characleristics. The L-band plots (Figurel b) show positive relationships. The highest correlation coefficient was derived for the W
polarization data (r=0.62, significant at the 95% level of confidence). Strongerrelationships may have been obtained with the longer wavelength P-band data.
3. BACKSCATTER RELATED TO STAND AGE.
Over large areas it ks logistically iml:x_sible to carry out intensive field surveys
in order to derive biomass estimates. However, accurate Forestry Commission stock maps
exist for the whole of the Tywi forest containing information regarding the planting dates
of forest stands. Stand age was used as a surrogate for biomass and only unthinned stands
18
"O
2.J
(a)0.8 0.8-
o.6- o.6-oo _. 0.4
•0.4"
0"2t
00
r = 0.21
i i n u u
5 10 15 20 25 30
Age (Years)
0.8
0.2"
o0
(e)
Ca)
r = 0.34
u u i
10 15 20 25
Age (Years)
30
E<
0.2-r :0.71
i _ ; u *
5 10 15 20 25 30
Age (Years)
Figure 2. Plots of L-band HH amplitude versus stand age for (a) all stands; (b) stands on
steep slopes facing away from the SAR removed; (c) stands on level ground removed.
were considered in the analysis ensuring, as far as possible, uniform canopy structure and
density of boles.
Figure 2a shows a plot of L-band HH polarization backscatter versus stand age.
"lhere was a general increase in backscatter until the age of 25 years. There was then
significant variability in backscatter illustrated by a weak negative relationship. However,it was to be expected that as trees reach maturity that the amount of backscatter will
increase at a slower rate until a saturation point is reached (Groom et al., 1992). It was
expected that topographic variations will influence the amount of backscatter. In order toreduce the effect of abnormally low backscatter, all stands at an aspect of between 90 ° and180 ° relative to the SAR look direction and with slope angles greater than 5 ° were
eliminated (Figure 2b), i.e. stands on steep slopes facing away from the sensor wereexcluded from the analysis. Some of the variability disappeared and the correlation rose
to r=0.34.
Much of the variability that still existed was found to be the backscatter from
those stands that were situated on flat ground (slope angles typically less than 10). These
stands were located in a basin surrounding a natural lake. Productivity of Sitka sprucestands is related to soil water content, which varies as a function of slope angle (Coutts
and Philipson, 1978). Stands situated on well drained slopes tend to have higher growthrates and hence more biomass compared to stands of a similar age in areas of poorly
drained soils. Therefore topography indirectly determined the amount of biomass
19
production.Whenstandssituatedonslopeanglesof less than 1° were eliminated from the
data set the resulting relationship (Figure 2c) resembled those derived by previous authors
from forests in different environments, and the r-value again increased (r=0.71). Thereforein this instance, age is not a good surrogate for biomass unless other environmentalinformation is taken into account.
4. CONCLUSION.
This study has shown preliminary results of analysis of radar backscatter in a
dense spruce forest plantation in an upland environment, although these conclusions needto be validated with larger data sets. It was found that:
1. A significant relationship exists between L-band VV backscatter and stem biomass
estimates derived from intensive field survey.
2. Topographic variation influences the amount of backseatter directly as a result of
viewing geometry and indirectly as a determinant of biomass production. This latter pointlimits the use of age as a surrogate for biomass.
5. ACKNOWLEDGEMENTS.
This research was funded by NERC studentship GT4?91flLS/62. I am grateful to the U.K.
Forestry Commission for supplying the ground data, the sponsors of MAC Europe for
supplying the AIRSAR data, the Commission of the European Communities for the
POLTOOL software, and Giles Foody for comments during the editing stages.
6. REFERENCES.
Coutts, M. P. and Philipson, M. P., 1978, "Tolerance of Tree Roots to Waterlogging. 1.
Survival of Sitka spruce and Lodgepole pine," New Phytology, vol. 80, pp. 63-69.
Curran, P. J. and S. Plummer, 1992, "Remote Sensing of Forest Productivity," NERCNews, vol. 20, pp. 22-23.
Groom, G. B., Mitchell, P. I.., Baker, J. R., Settle, J. J. and Cordey, R. A., 1992,
"Relationships of Backscatter to Physical Characteristics of Pine Stands in
Thetford Forest, U.K.," MAESTRO- 1/A GRISCA TT, Radar Techniques for Forestry
and Agricultural Applications. Final Workshop, ESA, ESTEC, Noordwijk, TheNetherlands, 1, pp. 137-141.
Kasischke, E. S. and Christensen, N. L., 1990, "Connecting Forest Ecosystem and
Microwave backscatter models," International Journal of Remote Sensing, Vol.11, no. 7, pp. 1277-1298.
Kasischke, E. S., Bourgeau-Chevez, L. L., Christensen, N. L., and Dobson, M. C., 1991,
"'Ihe Relationship Between Above-ground Biomass and Radar Backscatter as
Observed in Airborne SAR Imagery," Proceedings of the thirdAIRSAR workshop,
May 23-24, 1991, Pasadena, JPL Pub. 91-30, JPI.., Pasadena, pp. 11-21.l_eToan, T., Beaudoin. A., Riom, J. and Guyon, D., 1992, "Relating Forest Biomass to
SAR data," IEEE Transactions on Geoscience and Remote ,Sensing, Vol. 30, pp.403-411.
Luckmann, A.J., and Baker, J.R., 1993, "AIRSAR Observations of a Temperate Forest andModelling of Topographic Effects," Presented at the 25th International
Symposium, Remote Sensing and Global Environment Change, Graz, Austria, 4-8April 1993.
20
N95- 23944
FOREST INVESTIGATIONS BY POLARIMETRIC AIRSAR DATA
IN THE HARZ MOUNTAINS *
M. Keil, D. Poll, J. Raupenstrauch, T. Tares**, R. Winter
German Aerospace Research Establishment (DLR),
D-82234 Oberpfaffenhofcn, Germany
** Helsinki Univ. of Technology, Lab. of Space Technology,
SF-02150 Espoo, Finland
1. INTRODUCTION
The Harz Mountains in the North of Germany have been a study site lk)r several remote
sensing investigations since 1985, as tile ,nolmtainous area is one oI the forest regions m
Germany heavily affcctcd b 5' Iorest decline, especially in thc high altitudes above 800 m. In
a research programme at the University ol Berlin, methods are developed for improving
remote sensing assessment of forest structure and forest suite by additional (;IS
information, using several datasets for establishing a forest information system
(Kcnncweg, Schardt, Sagischcwsky, 1992).
The Harz has been defined as a testsite lk)r the SIR-C / X-SAR mission which is going to
deliver multi frequency and multipolarizational SAR data from orbit. In a pilote project led
by DLR-DFD, these data are to be investigated for forestry and ecology purposes. In a
preparing flight ca,npaign to the SIR-C / X-SAR mission, "MAC EUROPE 1991",
performed by NASA/JPL, an arca of about 12 km by 25 kin in the Northern Harz was
covered with multipolarizational AIRSAR data m the C-, L- and P-band, including the
Brocken, the highest mountain of the Harz, with an altitude of 1142 m.
The multiparameter AIRSAR data are investigated for their information content on the
forest state, regarding the following questions:
- information on forest stand parameters like forest types, age classes and crown density,
especially for the separation of deciduous and coniferous forest,
-information on the storm damages (since 1972) and the status of regeneration,
-information on the status of forest destruction because of forest decline,
-influence of topography, local incidence angle and soil moislurc on the SAR daub.
Within the project various methods and tools have been developed for the investigation of
multipolarimetric radar backscatter responses and for discrimination purposes, in order to
use the multipolarization information of the compressed Stokes matrix delivered by JPL.
2. AVAILABLE DATA
The AIRSAR scenes were flown at July 12th, 1991. Two profiles were heading North with
looking angles of 45 (leg. and 55 deg., two profiles heading South, with the same looking
21
angles.ThereforedataareavailablerepresentinganilluminationofthemountainsfromtheEastandfromtheWest,eachprofilecutintoNorthandSouthscene(Keiletal.,1993).
Paralleltothesurvey,70testareasintheupperHarzhavebeencheckedatthegroundforforesttypeandtreecomposition,ageclass,crowndensity,topographicfeaturesandgroundcover.ALANDSATTMscenefromJuly10th1991(twodaysbeforetheAIRSARflight)andadigitalelevationmodelisavailableforcomparison.WithinacooperationwiththeTechnicalUniversityof Berlin,infraredcolourphotoscanbeused.A largeprogressforinvestigationwasreachedbyshortwhenabout140polygonsfromairphotointerpretationof foreststateandfrom forestmanagementdatacouldbe integratedwithinthiscooperation.ThedataoverlayisbasedonageocodingofthreeAIRSARscenesperformedbyJohanneumResearch,Graz,includingterraincorrection.
3. TOOLSFORINVESTI(;ATIONOFPOLARIZATIONDATA
Inordertostudysignaturesofpolarimctricbackscatter,severalextensionsoftheavailablePOLTOOLsoftwarebyJPLwerefoundnecessary.Thus,forthesynthesisoflx)larimetricinformationoftargetareas,aninputmaskfilewascoupledwiththeSYNTHESISsoftwarebasedonarbitraryclosedpolygons(Tares,1993).
BesidesseveralothertoolsIor visualizationandslalistical evaluations, the following
representation was found very useful: The possible discrimination of two forest stands on
certain polarization states can not be checked by comparing mean backscatter amplitudes
alone. Thus, a deduction of standard deviation, assuming Gaussian distribution, was
performed for the marked forest reference areas and given polarization states directly fro,n
the Stokes matrix (Tares, 1993). As ellipticity seemed not to have a dominant influence lor
discriminalion, a two dimensional representation was used ,nainly: For co-polarization
and cross-polarization, the stand-averaged backscalter amplitude + standard deviation
can be presented in dependence of orientalion angle.
By known mean values M and standard deviation Sdev for two landcovcr classes a and b, an
expected accuracy of a Bayesian classificalion for one feature can be estimated using the
separability index (Dobson el al., 1992):
S (a,b/ = abs(M(a) M(b)) / (Sdcv(a) + Sdcv(b).
For Gaussian distribution, values of S > 1.5 correspond to a classification accuracy bctlcr
than 90%. The new developed program "discrimination" cqables a separability
optimization by changing both orientation anglcs and clliplicity angles for pairs of forest
classcs(Tares, 1993).
The statistical evaluations _crc used to prepare classification investigations. In order to
take into account the speckle and textural inlbrmalion, first classification approaches were
performed using the EBIS system (an "evidenced bascd inteq3remtion system") by
Lohmann, 1991, developed and installed al DLR. By this system there are supported
distribution functions describcd by, multinomial statistics besides Gaussian dislributions.
22
4. PRESENT RESUH'S
From visual interpretation and statistical evaluations, several characteristics could be
deduced for the AIRSAR data of the Harz study site.
L-band data seem to delivcr the best single band information for separation of deciduous
and conilerous stands, when two or three characteristic polarization states can be used. This
is due to the fact, that polarization response vary much more for spruce stands than li)r
deciduous stands with changing orientation angle (showing a maximum near HH and a
minimum near VV for co-polarization of spruce). Age classes arc difficult to delineate in
L-band alone, cultures and clearings can bc detected well (Raupcnstrauch, 1993). The
high-level spruce stands with Iowcr crown dcnsily can be distinguished against the dcnscr
stands in the North, for which the different texture gives also iuilX)rlant )hior|hal)on.
The three polarizations :it P band show thelughcstditlcrcnciation _.ilhinthe fores(areas.
Two influences besides crowll and stein paranieiers scent it) overlay Iorcsl illforlnation:
P-barid shows the slrongcsl rckilions to the It)pography, P-tttt and P VV, not P ttV, rcflecl
information not correlated with forest stand paranicters; investigations are planned to
check how far soil types, soil moisture and ground cover are responsible for thai.
Weakest influences by topography arc shown m the C-band. The polarization intkmnation
is smaller than in L-and P-band, leading to a low differentiation between deciduous and
coniferous trees. There is a higher potential for the differentiation of regeneration states
(clearings / cultures) and additional infornlation for ago class separation (thick(is / timber).
An example of polar)metric signatures for thrcc different stand types is shown in Fig. 1. The
spruce and the deciduous stand can be distinguished in L and P-band, not in C-band, the
orientations VV and HV seem best for that separation. VV-polarization has proved the
most informative orientation for C-band, e.g. for the separation of cultures from spruce,
pole and old timber (Kcil ctal., 1993).
For the present classifications by EBIS, dalasets of L-HH, L .VV and L HV as wcll as
C-VV and P-HV proved a suitable base. hnl_)rtant for classification was thc use of local
incidence infornialion which ix avadablc fiery in the co-rcgistialcd incidence ailglc ili_.isk.
5. REFERENCES
Dobson, M. C., L. Pierce, K. Saraband), F. T. Uhlby, T. Sharik, 1992, "Preliminary Analysis
of ERS-1 SAR for Forest Ecosystem Studies", IEEE Trans. on Geoscience and Remote
Sensing, Vol. 30, No. 2, pp. 203-211.
Keil, M., J. Raupenstrauch, T. Tares, P,. Winter, 1993, "Investigations of Forest Infonnation
Content of Polarimetric AIRSAR Data in the Harz Mounufins", 25th International
Symlx)sium, Remotc Sensing and Global Environmenlal Change (ERIM), 4-8 April 1993,
Graz, Austria (in press).
23
0.g
t0.6
J 0.4
i 0.2
P-BAND, LINEAR GO-POL, Ames: 725 48
• ... ....... ... ........ .. ...... ....'"
46 go 136 111)
Orientation Angle
L-BAND, LINEAR CO-POL, Areas: 7 25 48
o.g
0.8O4
<_ 02
i
o.e
i 0.4
1
. . , ...... , ...... , .....
0.0 - ....... , ......
.... ,6..... 80 ..... ,:. • ,:80Orientltion Angle
C-BAND, LINEAR CO-POL, Areas: 7 25 48
o.I
...................................-.:: _.
45 90 135 160
Orientation Angle
t°.'1
o.o?i0 48 90 135 180
Orientation Angte
L-BAND, LINEAR CROSS-PaL, A/eas: 7 25 48
O.I I ..... , ..... ' .... , .....
O.e
0.4
0.2
0.01 .......................... , ......
0 45 go 135 180
Oflento|_on A tl_ I Ip
C-BAND, LINEAR CROSS-PaL, Areas: 7 25 48
Oil ........ ' ........
t
!0,01 ............... , ...............
0 45 gO 135 180
Oriental}on Angle
Fig. 1. Polarization signatures for three stands: -- spruce, pole and old timber, mainly
closed, slope E 5 deg. - - - deciduous stand, old timber, slope SW, 20-30 dog.
.... culture (spruce), ()pen, 0 deg.
Kenneweg, H.,M. Schardt, H. Sagischewski,1992, "Observation of forest damages in the
Harz Mountains by Means of Satellite Remote Sensing", European "International Space
Year" Conference 1992, Munich, Germany, 30 March - 4 April 1992, ESA - ISY 1, Vol II,
pp. 805-810.
Lohmann, G., 1991, "An Evidential Reasoning Approach to the Classification of Satellite
Images", Diss. T.U. Miinchen, DLR-FB 91. _9,, Obcrpfaflcnhofen.
Raupenstrauch, J., 1993, "Untersuchungen zum Informationsgchalt von Multi-
polarisations- und Multifrequenz-SAR-Daten, insbesondere fiir forstliche Frage-
stelhmgen, unter Verwendung von anderen Fernerkundungs- und von Zusatzdaten", thesis
for diploma, Ludwig Maximilian University of Munich.
Tarcs,T., 1993, "AIRSAR Software - Programs lk)r Preprocessing of JPL Compressed
Polarimetric SAR Data", Internal Report, DLR, Oberpfaffenhofen.
24
N95- 23945
COMPARISON OF INVERSION MODELS USING AIRSAR DATADEATH VALLEY, CALIFORNIA
Kathryn S. Kierein-Young
Center for the Study of Earth from Space (CSES)
Cooperative Institute for Research in Environmental Sciences (CIRES)University of Colorado, Boulder, Colorado 80309-0449
and
Department of Geological SciencesUniversity of Colorado, Boulder, Colorado 80309
FOR
1. INTRODUCTION
Polarimetric Airborne Synthetic Aperture Radar (AIRSAR) data were collectedfor the Geologic Remote Sensing Field Experiment (GRSFE) over Death Valley,California, USA, in September 1989 (Evans and Arvidson, 1990; Arvidson et al, 1991).AIRSAR is a four-look, quad-polarization, three frequency instrument. It collectsmeasurements at C-band (5.66 cm), L-band (23.98 cm), and P-band (68.13 cm), and has aGIFOV of 10 meters and a swath width of 12 kilometers. Because the radar measures at
three wavelengths, different scales of surface roughness are measured. Also, dielectricconstants can be calculated from the data (Zebker et al, 1987).
The scene used in this study is in Death Valley, California and is located over
Trail Canyon alluvial fan, the valley floor, and Artists Drive alluvial fan. The fans arevery different in mineralogic makeup, size, and surface roughness. Trail Canyon fan islocated on the west side of the valley at the base of the Panamint Range and is a large fanwith older areas of desert pavement and younger active channels. The source for thematerial on southern part of the fan is mostly quartzites and there is an area of carbonatesource on the northern part of the fan. Artists Drive fan is located at the base of the BlackMountains on the east side of the valley and is a smaller, young fan with its source
mostly from volcanic rocks. The valley floor contains playa and salt deposits that rangefrom smooth to Devil's Golf Course type salt pinnacles (Hunt and Mabey, 1966).
2. CALIBRATION
The AIRSAR data were calibrated to allow extraction of accurate values of rms
surface roughness, dielectric constants, sigma-zero backscatter, and polarizationinformation. The data were calibrated in two ways, assuming a flat surface, and using a
digital elevation model to remove topographic effects. Both calibrations used in-scenetrihedral comer reflectors to remove cross-talk, and to calibrate the phase, amplitude, and
co-channel gain imbalance (van Zyl, 1990). The altitude of the aircraft was measuredincorrectly because the plane was flying over the Panamint mountains and imaging thevalley floor. This was corrected in the calibration in both cases. A digital elevationmodel (DEM) was generated by digitizing four USGS topographic quads using Arc/Info.This DEM was registered to the radar scene and used in the calibration to remove theeffects of topography (van Zyl et al, 1992). The near-range part of the image containsTrail Canyon fan which slopes away from the radar look direction and has the largesttopographic effect. Artists Drive fan, in the far range, has a gentle slope which did noteffect the calibration greatly. The comer reflectors used in the calibration were located onTrail Canyon fan. In the calibration without the DEM correction, the calibratedpolarization signatures for the comer reflectors were not ideal. However, once the DEMwas used, the calibrated signatures were much better. Figure 1 shows before and after
DEM calibration polarization signatures for a comer reflector in C-band. Also, thesigma-zero values changed slightly with the DEM correction. Areas that face away from
25
theradarlook direction due to topography have DEM corrected sigma-zero values that aregreater than (less negative) those that are DEM uncorrected. Areas that face toward the
radar look direction have DEM corrected sigma-zero values that are less than (morenegative) those that are DEM uncorrected. Areas that are flat, without topography, havesigma-zero values that are the same in both calibrations.
3. INVERSION AND ANALYSIS
The first-order small perturbation model (Evans et al, 1992; van Zyl et al ,1991;Barrick and Peake, 1967) was used to estimate the surface power spectral density and thedielectric constant at every pixel by performing an inversion using the AIRSAR dam.This model is valid only for very smooth surfaces. Results from the small perturbationmodel inversion are three values, one for each of the radar frequencies, that describe thepower spectral density of the surface and a value for the dielectric constant at eachfrequency. The power specmim of a geologic surface is approximately linear in log-logspace. Fitting the three points from the inversion with a line using a least-squaresmethod produces slope and intercept values that allow calculation of the fractal dimensionof the surface and arms surface roughness value. The slope of the power spectrum isrelated to the two-dimensional fractal dimension of the surface. The fractal dimension of a
surface describes the scaling properties of the topography (Mandelbrot, 1982). A surfacemay have a fractal dimension between 2 and 3 and as the fractal dimension increases,heights of nearby points become more independent (Brown and Scholz, 1985). Theintercept of the power spectrum can be directly related to a rms surface roughness usingforward modelling. Using the fractal dimension and rrns surface roughness calculatedfrom the radar inversion power spectrum, a synthetic three dimensional plot can be madethat represents the surface (Huang and Turcotte, 1989; Kierein-Young and Kruse, 1992).
A modified small perturbation model (van Zyl, personal communication; Zebkeret al, 1991) was also used to estimate the rms surface height and dielectric constant atevery pixel by performing an inversion using the AIRSAR data. This model modifies the
small perturbation model to extend the validity range to all surfaces by including anempirically derived function that approximates the change in roughness with backscatter.Two versions of the modified small perturbation model inversion were used. The first
assumes a constant value for the slope of the power spectrum. The second method uses,at every pixel, the power spectrum slope obtained from the small perturbation modelinversion. Results from both the modified small perturbation models are values for therms surface roughness and dielectric constant for each frequency.
4. RESULTS AND CONCLUSIONS
The results from the three inversion models are shown in Table 1 compared withfield data for five sites. These five sites consist of three alluvial fan units, a playa, andDevil's Golf Course type salt pinnacles. The active fan site is from the most activechannel on Trail Canyon fan, the desert pavement site is also from Trail Canyon fan, andthe third site is from Artists Drive fan. Field surface roughness data were obtained fromdigitized helicopter stereo pairs (Farr, personal communication) of each site and from anUSGS Open File Report (Schaber and Berlin, 1993). Field dielectric measurements weremade with a C-Band dielectric probe. In general, the results from the small perturbationmodel underestimates the value of surface roughness. In both of the modified smallperturbation models, the surface roughness values generally increase with frequency. Theroughness values in the model with a constant power spectrum slope value (y) are largerthan those from the model using the power spectrum slope from the small perturbationmodel. The P-Band data seem to match the field data more closely than the other bands.However, this may be because of the dielectric values. The dielectric values are mostreasonable in the small perturbation model. The dielectric values in the modified models
are too high except in P-Band. The modified small perturbation model was triedassuming a dielectric constant equal to 3.0 for every pixel. This inversion producedsurface roughness values that were too large in most cases.
26
Usinginversionmodelstoobtainsurfaceroughnessand dielectric constants fromAIRSAR data produces quantitative results that are reasonably accurate. The smallperturbation model tends to give the overall best results for dielectric constants. C-bandand L-band data in the modified small perturbation model tend to produce dielectricconstants that are too high. P-band data in the modified model produces the best overallresults for both surface roughness and dielectric constant. The results from the modifiedmodel with a constant power spectrum slope are similar to those of the model withvariable power spectrum slopes. Since surfaces do show differences in their powerspectrum slopes, it is more accurate to include this variation in the inversion model. Theresults of these inversion models can be used to help in determining the active surficial
processes and the age of alluvial fan surfaces. Combining the inversion results with datafrom other sensors can help to determine why roughness characteristics are spatiallyvariant (Kierein-Young and Kruse, 1991). For example, the north part of Trail Canyonfan has a lower surface roughness than the rest of the fan. This difference is due to a
different, more easily eroded source material.
5. ACKNOWLEDGEMENTS
Portions of this work were supported under a NASA Graduate Student Researchers GrantNGT-50728. Portions of this work were supported under the SIR-C project byNASA/JPL contract 958456. Field work was supported by a grant from the Departmentof Geological Sciences, University of Colorado.
6. REFERENCES
Arvidson et al., 1991. GRSFE CD-ROM and documentation, version 1:USA_NASA_PDS GR 0001, Planetary Data System, NASA (9 CD-ROMs).
Barrick, D. E., and W. H. Peake, 1967. Scattering form surfaces with different roughnessscales: Analysis and interpretation, Research Rept. No. BAT-197A-10-3, BattelleMemorial Institute, Columbus Laboratories, Columbus, Ohio, AD 662751.
Evans, D. L., and Arvidson, R. E., 1990. The Geologic Remote Sensing FieldExperiment (GRSFE): Overview of initial results: in Proceedings, IGARSS'90, University of Maryland, College Park, Md., The Institute of Electrical and
Electronics Engineers, Inc. (IEEE), New York, p. 1347.Evans, D. L., T. G. Farr, and J.J. van Zyl, 1992. Estimates of surface roughness derived
from synthetic aperture radar (SAR) data, IEEE Transactions on Geoscience andRemote Sensing, V. 30, No. 2, pp. 382-389.
Huang, J. and D. L. Turcotte, 1989. Fractal mapping of digitized images: Application tothe topography of Arizona and comparisons with synthetic images, Journal ofGeophysical Research, V. 94, No. B6, pp. 7491-7495.
Hunt, C. B. and D. R. Mabey, 1966. Stratigraphy and structure in Death Valley,California. U. S. Geological Survey Profession Paper 494-A, 162p.
Kierein-Young, K. S.and F. A. Kruse, 1992. Extraction of quantitative surfacecharacteristics from AIRSAR data for Death Valley, California, in Summaries ofthe third annual JPL Airborne Geoscience Workshop, V. 3, pp. 46-48.
Kierein-Young, K. S. and F. A. Kruse, 1991. Quantitative Investigations of GeologicSurfaces utilizing Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) andPolarimetric Radar (AIRSAR) Data for Death Valley, California, in Proceedingsof the Eighth Thematic Conference on Geologic Remote Sensing, V. 1, pp. 495-506.
Mandelbrot, B. B., 1982. The Fractal Geometry of Nature, W. H. Freeman, SanFrancisco, California.
Schaber, G. G. and G. L. Berlin, 1993. Death Valley, California: Surface micro-reliefstatistics and radar scatterometer data, U. S. Geological Survey Open-file Report
93-272, 43p.van Zyl, Jakob J., 1990. Calibration of Polarimetric Radar Images Using Only Image
Parameters and Trihedral Comer Reflector Responses, IEEE Transactions onGeoscience and Remote Sensing, V. 28, No. 3, pp.337-348.
27
vanZyl,JakobJ.,Charles F. Burnette and Tom G. Farr, 1991. Inference of surface
power spectra from inversion of multifrequency polarimetric radar data,Geophysical Research Letters, V. 18, No. 9, pp. 1787-1790.
van Zyl, J. J., B.D. Chapman, P. C. Dubois, and A. Freeman, 1992. Polcal User'sManual, .rPL D-7715 version 4.0.
Zebker, Howard A., Jakob J. van Zyl, and Daniel N. Held, 1987. Imaging RadarPolarimetry From Wave Synthesis, Journal of Geophysical Research, V. 92,No. B1, pp. 683-701.
Zebker, H. A., J. J. van Zyl, S. L. Durden, and L. Norikane, 1991. Calibrated imagingradar polarimetry: Technique, examples, and applications, IEEE Transactions onGeoscience and Remote Sensing, V. 29, No. 6, pp. 942-961.
2O
Co_POlarized [C] Co_Polarized [C]
Figure 1. Co-polarized C-band comer reflector polarization signatures without DEM correction (left) and withDEM correction (right).
Table 1. AIRSAR inversion model results.
SPMSurface RM$ f_l
Active Fan C 6.5 2.13 4.0L 3.6P 3.0
Desert PvtFan C 2.0 2.18 10.3L 8.8P 3.0
Artists Dr. Fan C 1.0 2.405 2.4L 2.3P 2.7
Playa C 2.3 2.105 4.8L 10.9P 2.5
Devil's Golf Cs. C 19.0 2.03 3.0L 2.5P 3.0
MSPM y=2.55 MSPM y=SPM FieldRM$ (t RM$ ¢ RM$2.7 66.8 2.1 68.4 17.4 3.16.7 27.3 5.0 27.3
10.9 4.2 10.3 3.61.4 26.9 1.2 25.7 9.8 3.02.7 12.3 2.6 10.8
10.7 1.5 6.9 1.4
2.6 46.1 2.6 28.7 11.5 2.82.5 10.0 3.4 13.69.4 2.1 11.8 4.31.5 28.3 1.0 28.3 7.6 5.30.9 26.1 0.5 20.86.2 3.9 9.9 3.3
NGP NGP NGP NGP 12.0 2.412.1 50.7 5.7 27.530.3 17.7 28.2 39.4
RMS = root mean square surface roughness in cmfd = fractal dimensione ---dielectric constant
NGP = no good points
28
N95. 23946
GEOLOGIC MAPPING USING INTEGRATED
AIRSAR, AVIRIS, AND TIMS DATA
Fred A. Kruse
Center for the Study of Earth from Space (CSES) and Departmentof Geological Sciences, University of Colorado, Boulder 80309-0449
I. INTRODUCTIONThe multi-sensor aircraft campaign called the "Geologic Remote Sensing Field
Experiment" (GRSFE), conducted during 1989 in the southwestern United States,collected multiple airborne remote sensing data sets and associated field and laboratorymeasurements (Arvidson et al., 1991). The GRSFE airborne data sets used in this studyinclude the airborne Synthetic Aperture Radar (AIRSAR), the Airborne Visible/InfraredImaging Spectrometer (AVIRIS), and the Thermal Infrared Multispectral Scanner fillS)(See Table 1). Each sensor's unique characteristics were used for this study in a combined
analysis scheme for geologic mapping. AIRSAR was used to map structures andlandforms, AVIRIS was used to map mineralogy, and TIMS was used to map lithology.Visual data integration using IHS transforms and combined numerical analysis usingderived geophysical and geologic parameters with "multispectral" techniques resulted inimproved geologic mapping over that possible using each data set individually.
TABLE 1.
Sensor
AIRSAR
AVIRIS
_TIMS
GRSFE Sensor Specifications
Wavelength Number of
range channels
Pr L, C bands 3 (Quad pol)
68.13, 23.98 r 5.66 cm
0.41 - 2.45 ktm 2248.2 - 11.7 lam 6
Sampling SpatialInterval Resolution
NA 6-12 m
10 nm 20 m
400- 800 nm 20 m
II. LOCATION AND GEOLOGYThe extreme northern end of Death Valley (northern Grapevine Mountains)
California and Nevada, USA, has been studied in detail using field dam and several remote
sensing data sets (Kruse, 1988, Kruse et al., 1993). Bedrock in the area consists oflimestones, dolomites, sandstones and their metamorphic equivalents along with quartz
syenite, a quartz monzonite porphyry stock, and quartz monzonite dikes (Wrucke et al,1984, Kruse, 1988). The igneous rocks are cut by narrow north-trending mineralized
shear zones and slightly broader northwest-trending zones of disseminated quartz, pyrite,sericite, chalcopyrite, and fluorite mineralization (QSP alteration) as well as several smallareas of quartz stockwork. Skarn, composed mainly of brown andradite garnet intergrownwith calcite, epidote, and tremolite, occurs around the perimeter of the quartz monzonitestock. Tertiary volcanic rocks are abundant around the periphery of the study area.Quaternary deposits include Holocene and Pleistocene fangiomerates, pediment gravels,and alluvium.
III. AIRSAR DATAThe JPL P-, L-, and C-band airborne imaging radar polarimeter data (AIRSAR)
were calibrated using Trihedral comer reflectors placed during the GRSFE experiment.The comer reflectors were not physically in the northern Grapevine Mountains scene, butin an adjacent scene in the same flightline. Calibration consisted of calibration of phase,cross-talk, co-channel imbalance, and absolute radiometry using software developed at JPL(van Zyl et al., 1992).
The AIRSAR data were converted to radar backscattering coefficient (a °) to
allow extraction of backscatter signatures for areas indicated as having varying surface
29
roughnessonaPLC(RGB)colorcompositeimage.SelectedbackscattersignaturesfromthenortbernGrapevineMountainssiteareshowninFigure1.Thebackscattersignaturesshowthatingeneralthoserocksandfansthatcontainalteredmaterialaresmootherthanthosethatdonot.Thisestablishesalinkbetweensurfacemorphologyandcompositionfortheseparticularunits.Somefansofdifferingcomposition(volcanicfansandmixedcompositiongraniticfans)withsimilarspatialrelationstounalteredbedrock,however,havesimilarbackscattersignatures.ArockfaceperpendiculartothelookdirectionoftheAIRSARacts,asexpected,nearlyasacomerreflector.
Figure1. Radarbackscattersignatures (o °) for selected rock units.
0 j _L ROUGHC-BAND L ;AND
-- " P-BAND
_ Z
i -20 z
o-j
o
•.30_ ROCK FACE -__,..._
PROXIMAL FAN i
erVOLCANIC FAN
"40 MIXED COMPOSITION GRANrTIC FAN
FAN FROM ALTERED AREA
ALTERED QI"Z MONZOMTE PORPHYRY-50 , SMOOTH
0 2o 4o 6o 80WAVELENGTH (cm)
The AIRSAR data were also inverted to rms surface roughness and fractaldimension using a first-order small perturbation model (Kierein-Young, 1993). Thesurface roughness was calculated from the intercept of a log-log plot of the power spectraldensity for the PLC data and the fractal dimension from the slope. The distribution ofsmooth versus rough areas generally corresponded to that observed during the interactiveextraction of frequency signatures above. Again, the altered areas were shown to berelatively smooth. Density sliced images of the rms roughness, however, allowedquantitative spatial mapping of several ranges of roughness.
IV. AVIRIS DATA
AVIRIS is an imaging spectrometer measuring reflected light in 224 narrow (10m-wide) bands between approximately 0.4 -2.45 I.tm (Porter and Enmark, 1987).Numerous studies have used these data for mineralogic identification and mapping based
on the presence of diagnostic spectral features. Field spectra of known targets (one lighLone dark) were used to calibrate the AVIRIS data to reflectance with the empirical line
method (Kruse et al., 1993). Spectral signatures were extracted using interactive displayand analysis software allowing identification of individual minerals and mineral mixtures(Kruse et al., 1993). An expert system was used to identify spectral endmembers and
linear spectral unmixing was used to produce quantitative image-maps showing theirspatial distribution. Thematic mineral maps provide detailed surface compositionalinformation regarding secondary minerals (alteration and weathering). The resultsillustrate the fracture-controlled nature of most of the alteration.
V. TIMS DATA
TIMS is an aircraft scanner measuring six channels in the 8-12 lam range(Palluconi and Meeks, 1985). Silicate minerals have diagnostic emissivity minima inthis spectral region (Lyon, 1965). TIMS data acquired at 13:22 local time were calibrated
to radiance using gain and offset values for every scanline calculated using Planck'sradiation law with the TIMS internal reference values and channel spectral responses. "Iheradiance values were next converted to emissivity using the "model emittance" or "black
body" technique described by Kahle (1987), assuming that the emissivity in TIMS band 6was equal to 0.93. No atmospheric correction was made, however, examination of
individual TIMS spectra, indicates that the emissivity calibration was adequate without
3O
the allnospheric correction, probably because of the extremely arid nature of the studyare_
Color composites were made from the calibrated TIMS bands and emissivity
spectra were extracted from the calibrated data for areas with known rock compositions.The principal feature of these spectra is the presence of a subtle emissivity minimum inTIMS band 3 (9.18 lam) caused by the fundamental silicon-oxygen stretching vibrationsof quartz and other silicate minerals. Although the emissivity features are subdued in theTIMS spectra, the shift in the minimum from TIMS band 3 to TIMS band 4 associatedwith decreased silica content can be observed for several silica-poor rock units. When 4/3,
5/4, and 3/2 (RGB) emissivity ratios are used together in a color-ratio-composite image,the silica-rich areas are obvious red areas on the image, while the areas with less silica
appear green to blue-green. The designed ratios and image colors were used to produce
lithological maps.
VI. IMAGE INTEGRATION FOR VISUAL INTERPRETATION
All of the data sets were co-registered to a map base using a DEM.Combinations of the diverse data sets were then made using the Intensity, Hue, Saturation
(IHS) transform (Gillespie et al., 1986). In this procedure, the color image to be overlainis transformed from the RGB coordinate system into the IHS system, the intensity is
replaced by a single-band base image, and the new IHS coordinates are transformed back toRGB. This preserves the colors from the original color image while allowing use of thebase image to modulate the brightness. Several image combinations were made thatenhanced the usefulness of the data for geologic mapping. The geology and alterationmaps produced in the field were overlain on the radar data. These images allowedverification of the field mapping. Additionally, the alteration overlay on the radar datashowed previously unrecognized structural control of alteration. TIMS data overlain onL-band AIRSAR showed additional associations of lithology and structure. The most
useful image utilizing combined radar and optical remote sensing, however, was thecombination of the P-, L-, C-band (RGB) image with the AVIRIS mapping results.
These images and the o ° signatures show that roughness information is linked tocomposition obtained from the optical remote sensing. For example, the areas thatappear smooth at P- and L-band and rough at C-band are judged to have 5 cm-scale surface
relief. The AVIRIS analysis indicates that these areas also have abundant alterationminerals (sericite). One interpretation of these associations might be that the altered areashave reduced the surface relief relative to surrounding rocks. This is supported by fieldwork, which shows that these areas correspond to highly altered rocks that have been
weathered to a desert-pavement-like surface.
VII. COMBINED ANALYSIS OF DERIVED DATA SETSThe coregistered derivative data sets were combined into a "data cube" for
statistical and multispectral analysis. Derived data sets used included surface roughnessand ffactal dimension from the AIRSAR data, mineral abundance images from the
AVIRIS data, and emissivity ratio images from the TIMS data. "Spectral" and supervisedclassification methods were used on the combined data set to produce image maps
showing broad geomorphic units with common roughness, lithological, and alterationcharacteristics. These images support the previous link between surface roughness andcomposition while providing spatial maps showing the associations and distribution.
VIII. CONCLUSIONS
The integrated multispectral images are providing new geologic information thatcan be used for lithologic and structural mapping to assist development of geologicmodels. The AIRSAR data provide information about the surface morphology of rocksand soils and the scale of surface roughness. They allow definition of geomorphic unitswhich indicate that there is a link between surface morphology and composition for someunits. The AVIRIS data allowed direct compositional mapping of surface mineralogy.The TIMS extended spectral signatures provide compositional information not containedin the visible, near-infrared or short wave infrared data. The TIMS spectra and imageswere useful for mapping lithoiogical differences between igneous rock types because
31
TIMS is particularly sensitive to silica content. They also showed several areas of veryhigh silica concentrations corresponding to silicified (altered) outcrop. The integratedmultispectral images provide a case study of the characteristics of these diverse data sets
and their capabilities for geologic mapping. The combined analyses of these datademonstrates that improved lithological, mineralogical, and structural mapping ispossible using remote sensing, even where extensive, detailed ground mapping has beencompleted.
The next step in this study is to use the GRSFE and other data sets to extend theanalysis to a regional scale. Extension of the study area to include much of the northernDeath Valley region is in progress. SIR-C data will be used in conjunction with AVIRISand TIMS to produce detailed geologic maps that will allow development of regionalgeologic models for the evolution of igneous, metamorphic, and sedimentary rocks andrecent geologic surfaces.
IX. ACKNOWLEDGEMENTS
Portions of this work were supported by NASA Grants NAGW-1143 and
NAGW- 1601. Additional support for analysis of AIRSAR data was provided byNASAJJPL grant 958456.
X. REFERENCESArvidson et al., 1991, GRSFE CD-ROM and documentation_ version 1:
USA_NASA_PDS GR 0001, Planetary Data System, NASA (9 CD-ROMs).Giilespie, A. R., Kahle, A. B., and Walker, R. E., 1986, Color enhancement of highly
correlated images. I. De.correlation and HSI contrast stretches: Remote Sensingof Environment, v. 20, p. 209-235.
Kahle, A. B., 1987, Surface emittance, temperature, and thermal inertia derived fromThermal Infrared Multispectral Scanner (TIMS) data for Death Valley, California,_, v 52, no. 7, p. 858-874.
Kierein-Young, K. S., 1993, Comparison of inversion models using AIRSAR data forDeath Valley, California, in Summaries. 4th Airborne Geoscience W0rk_h0p,
25-29 October, 1993, Washington, D. C., (in press).Kruse, F. A., 1988, Use of Airborne Imaging Spectrometer data to map minerals
associated with hydrothermally altered rocks in the northern Grapevine Mountains,Nevada and California: Remote Sensing of Environment, V. 24, No. 1, p. 31-51.
Kruse, F. A., Lefkoff, A. B., and Dietz, J. B., 1993, Expert System-Based MineralMapping in northern Death Valley, California/Nevada using the AirborneVisible/Infrared Imaging Spectrometer (AVIRIS): Remote Sensing of
_, Special issue on AVIRIS, May-June 1993, p. 309 - 336.
Lyon, R. J. P., 1965, Analysis of rocks by spectral infrared emission (8 to 25 microns):Economic Geology. v. 60, p. 715-736.
Palluconi, F. D. and G. R. Meeks, 1985, Thermal Infrared Multispectral Scanner(TIMS): An investigator's guide to TIMS data, JPL Publication 85-32. Jet
Propulsion Laboratory, Pasadena, CA, 14 p.
Porter, W. M., and Enmark, H. T., 1987, A system overview of the AirborneVisible/Infrared Imaging Spectrometer (AVIRIS): in Proceedings. 31st Annual
International Technical Symposium. Society of Photo-Optical InstrumentationEngineers, v. 834, p. 22-31.
Wrucke, C. T., Werschky, R. S., Raines, G. L., Blakely, R. J., Hoover, D. B., and
Miller, M. S., 1984, Mineral resources and mineral resource potential of theLittle Sand Spring Wilderness Study Area, Inyo County, California: US.
Geological Survey Ope_n File Report 84-557, 20 p.
van Zyl, J. J., Chapman, B. D., Dubois, P. C., and, Freeman, A., 1992, POLCAL
User's Manual. Version 4.0. JPL D-7715. Jet Propulsion Laboratory, Pasadena,CA.
32
N95- 23947
STATISTICS OF MULTI-LOOK AIRSAR IMAGERY:
A COMPARISON OF THEORY WITH MEASUREMENTS
J.S. Lee, K.W. Hoppel and S.A. Mango
Remote Sensing Division, Code 7230
Naval Research Laboratory
Washington DC 20375-5351
1. INTRODUC'FION
The intensity and amplitude statistics of SAR images, such as L-Band HH forSEASAT and SIR-B, and C-Band VV for ERS-l have been extensively investigated
for various terrain, ground cover and ocean surfaces. Less well-known are thestatistics between multiple channels of polarimetric or interferometric SARs,
especially for the multi-look processed data. In this paper, we investigate the
probability density functions (PDFs) of phase differences, the magnitude of complex
products and the amplitude ratios, between polarization channels (i.e. HH,HV, and
VV) using l-look and 4-look A1RSAR polarimetric data. Measured histograms are
compared with theoretical PDFs which were recently derived based on a complex
Gaussian model (Lee et al., 1993).
NASA/JPL l-look and 4-look AIRSAR data of Howland Forest and SanFrancisco were used for comparison. Histograms from l-look SAR data agreed with
theoretical PDFs. However, discrepancies were found when matching the 4-look
polarimetric data with the 4-look PDFs. Instead, We found that the 3-look PDFsmatched better. The problem was traced to the averaging of correlated 1 -look pixels.We also verified these theoretical PDFs for forest, ocean surfaces, park areas and city
blocks. This indicates that ground cover and terrain with different scattering
mechanisms can be represented with this statistical model.
2. THE COMPLEX GAUSSIAN MODEL
A polarimetric radar measures the complete scattering matrix S of a medium
at a given incidence angle. For a reciprocal medium, the three unique complex
elements are Shh, Sh,,, and S_,. Circular Gaussian Conditions (Goodman, 1985) areassumed. The circular Gaussian assumption has been verified by Sarabandi(1992)
using l-look polarimetric SAR data.
The multi-look AIRSAR processor compresses polarimetric data by averaging
Mueller matrices of I-look pixels in the azimuth direction. Due to oversampling, the
neighboring 1-look pixels of AIRSAR data are somewhat correlated. For mathematical
simplicity, the PDFs for multi-look phase difference, etc. were derived under the
assumption of statistical independence.
3. MULTI-LOOK PHASE DIFFERENCE DISTRIBUTION
Let n be the number of looks, and Sz(k) and S2(k) be the kth l-look samples
of any two components of the scattering matrix. For polarimetric and interferometric
radars, the multi-look phase difference is obtained by
33
n1
=Arg[ n _[_ S, (k) S; (k) ] (Z)k=l
Under the circular Gaussian assumption, the multi-look phase distribution
were derived by multiple integrations of special functions (Lee et al., 1993). Themulti-look phase difference PDF is
p(@) = F(n+i/2) (m-lpcl=)n13+ (m-]Pcl=)n2vl-_ F(n) (1-[32) n'I/2 2re
F(n, i; 1/2; 132)(2)
with
13 = IPcl cos(_-O), -=<,1,<= (3)
where F(n, 1; I/2; f12) is a Gauss hypergeometric function, and 0 and Ipcl are the phaseand the magnitude of the complex correlation coefficient defined as
E[SzS2]P_= =lpcle i° (4)
(EtlS_I2]EtlS=I =]
The l-look (n=l) PDF can be obtained by applying mathematical identities ofthe hypergeometrical function (Lee et al., 1993). The l-look PDF obtained is identical
to that of Kong (1988) and Sarabandi (1992). The PDF of Eq. (2) depends only on thenumber of looks and the complex correlation coefficient. The peak of the distribution
is located at _b=8. A plot of standard deviation versus IpJ is given in Fig. 1, whichverifies a similar but less accurate figure (Zebker, 1992) computed with
interferometric radar data. As shown in Fig. 1, multi-look processing effectivelyreduces the phase error, especially when n=16 and 32.
4. DISTRIBUTION OF THE MAG NITU DE OF THE NORMALIZ ED MULTI- LOOKCOMPLEX PRODUCT
The magnitude of product of S 1 and S 2 is an important measure inpolarimetric SAR, and it is the magnitude of interferogram from an interferometricSAR. The normalized magnitude is defined as
n
1! r, (k)s;(k>l/2 k=l
_/E[IS_r "_] E[ IS212]
(s)
The PDF of _ (Lee et a1.,1993) is
p (F.) 4 r2n+z_ n 2 Ip_In_ 2n_= Io ( ) Kn_ 1 ( ) (6)r(.) i-I , I 1-1pcl=
where Io( ) and Kn( ) are modified Bessel functions. The PDF for the unnormalized
magnitude can be easily obtained from Eq.(6) by using Eq. (5). The standard
34
deviationsof thisPDFisplottedversusthecorrelationcoefficientin Fig.2. Thisplotindicatesthat the magnitudeof complexproductcanbe usedasa discriminator,especiallyfor largen.
5. MULTI-LOOK AMPLITUDE RATIO DISTRIBUTIONS
The amplitude ratio between Shh and Sv,, has been an important discriminator
in the study ofpolarimetric radar returns• Let the normalized ratio
n n
(_ ls_(k) !_/E[IS_12])/ (_ Is2(k)12/E[Is21_])k=l k=l
(7)
The multi-look amplitude ratio PDF (Lee et al., 1993) is
p(v) 2F(2n) (i-[_c12) n (l +v2 ) v2n-i= , 0 _v<oo (8)
F(n) F(n) [(l-v2)2-4]Pc12V2] n+I/2
For n=l, we have the l-look amplitude distribution, which is identical to the results
of Kong(1988), when his PDF is properly normalized.
2.0
._- 1.5
g_- 1.0>
121
_ o.5
O9
Number OI LOOKS:
\8
o,o .................. 0', • "0.0 0.2 014 0.6 8 1.0
Correlation Coefficient, Ip, I
Fig. 1 Phase difference standard deviations
versus the correlation coefficient, Ip¢l, forthe number of look from n= 1, 2, 4, 8, 16and 32. The value of 0=0.
g
0.8 ', Numlz,e, Of Looks /
o.6! _ i
0.2_: .... / ,6 .... -....
0o[ ................. ]0.0 0.2 0.4 0.6 0.8 1.0
Correlation Coefficient, b,I
Fig. 2 Standard deviations of Normalized
magnitude of multi-look interferogram
versus correlation coefficient, IpJ, for thenumber of look fromn= 1,2,4, Sand 16.
6. COMPARISON OF THEORETICAL PDFs WITH AIRSAR MEASUREMEN'IS
In this section, the A1RSAR polarimetric data was used to compute histograms
of phase differences, normalized products and amplitude ratios to verify this complexGaussian model• Homogeneous regions of forest were selected, and the complex
correlation coefficient (i.e., 0 and IpJ) and histograms were computed. We havechecked the 1 -look C-Band and L-Band data, and found that the agreements between
histograms and their corresponding PDFs are good. However, discrepancies exist inthe 4-look data. Using 4-look Howland Forest data (CM1084), the match between the
histogram and the 4-look phase difference PDF are not as good (Fig. 3A). A bettermatch was found with a 3-look PDF (Fig. 3B). Histograms of amplitude and
normalized product are also shown good agreements with 3-look PDFs. To save space,
only the case of HH to VV ratio is shown in Fig. 3C.
35
Theproblemwas traced to the averaging of correlated l-look neighboringpixels during the multi-look processing. To verify it, we use l-look data of Howland
Forest (HR1084C) and perform the 4-look processing by averaging four pixels
separated by two pixels in the azimuth direction. Since the correlation between everyother pixels is much less than that between its immediate neighbors, statistical
independence is assured. The results are shown in Fig.4 for all three variables under
study. The agreement with 4-look PDFs is very good. We can conclude that due to
the correlation of l-look data, the 4-look AIRSAR data has the characteristics of a3-look.
HH-VV Phase Difference H_ram
4° I o c.-aa,xl ::
-'>,> !30 1 °°
4-i.,,_P_ _o :.i _ °t * tJ\ 7
4°i HH-VV.pha_se [::)ifferenceHistogram _
o % C-l_incl
°<,i>°
30 oo
'l .jo;10
c+¢
....
HH-VV Phase Difference Histogram
=_o i i
I ,>_<>o
t
,oL :..'t."'°E201'-
.. o ° " _ i"
oL.,
HHNV Normalized Amplitude Ratio Histoo=.ram
-180 -135 -90 -45 0 45 IN) 135 180 -le0 -135 -90 _ 0 45 90 t35 180 0 1 2 3 4 $
Pl'umse Dd_ence (deg_ee_) Phl_, Odfefen_ {degree_) Ratio
Fig. 3 Experimental results using 4-look AIRSAR data of Howland Forest (CMI804)with ]pc[=0.491 and 0=84.9 °. (A) The histogram of phase difference between HH and
VV and the theoretical 4-look PDF. (B) The same histogram but with a 3-look PDF.
The match is better. (C) The histogram of normalized amplitude ratio, HH/VV andthe 3-look PDF.
HH*W Normalized Product Histocjram HH]VV Normalized Amplitude Ratio Histo_
• C_B,ancl C-Band
"o !
t N ,-<oo,°<>, :
"+<a"-- '° i20 t" _,.,r_-._ o
<,-;_._..., : 2o: ¢ o_ :
-180 -135 -IO -.45 0 45 90 135 180 0.0 05 1.0 1.5 2.0 25 0 1 2 3
Phase Diff_lrt_ (degre._) Product Ritlo
Fig. 4 Using l-look AIRSAR data (HRI084C), the 4-look data were computed byaveraging pixeis separated by a distance of two pixels. (A) HH-VV phase difference
histogram and the 4-look PDF. The agreement is good. (B) The histogram of
normalized IHH*VVI and the 4-look PDF. (C) The histogram of normalized IHHt/IVVIand the 4-look PDF.
4 $
REFERENCES
[l] Goodman, J.W., 1985, Statistical Optics, John Wiley and Sons, New York, 1985.
[2] Kong, J.A., et a]., 1988, " Identification of Terrain Cover Using the Optimal
Polarimetric Classifier," J. Electro. Waves Applic., 2(2) 171-194.
[3] Lee, J.S., K.W. Koppel, S.A. Mango and A.R. Miller, 1993,"Intensity and Phase
Statistics of Multi-look Polarimetric SAR Imagery," Proceedings of IGARSS'93.
[4] Sarabandi, K., 1992, "Derivations of Phase Statistics from the Mueiler Matrix",Radio Science, 27(5), 553-560.
[5] Zebker, H. A., and Villasenor, J.,1992, "Decorrelations in Interferometric Radar
Echoes," IEEE Geoscience and Remote Sensing, vol.30, no. 5,950-959.
36
N95- 23948
AIRSAR DEPLOYMENT IN AUSTRALIA, SEPTEMBER 1993:
MANAGEMENT AND OBJECTIVES
A. K. Milne
Centre for Remote Sensing and GIS
University of New South WalesPO Box 1
Kensington, NSW 2033
I. J. TapleyCSIRO Division of
Exploration and Mining
Private Bag, P.O.
Wembley, WA 6014
1. INTRODUCTION
Past co-operation between the NASA Earth science and ApplicationsDivision and the CSIRO and Australian university researchers has led to a number
of mutually beneficial activities. These include the deployment of the C-130
aircraft with TIMS, AIS, and NS001 sensors in Australia in 1985; collaborationbetween scientists from the USA and Australia in soils research whcih has
extended for the past decade; and in the development of imaging spectroscopy
where CSIRO and NASA have worked closely together and regularly exchanged
visiting scientists. In may this year TIMS was flown in eastern Australia on board
a CSIRO-owned aircraft together with a CSIRO-designed CO2 laser spectrometer.
The Science Investigation Team for the Shuttle Imaging Radar (SIRC-C)
Program includes one Australian Principal Investigator and ten Australian co-
investigators who will work on nine projects related to studying land and near-
shore surfaces after the Shuttle flight scheduled for April 1994.
This long-term continued joint collaboration was progressed further with
the deployment of AIRSAR downunder in September 1993. During a five week
period, the DC-8 aircraft flew in all Australian states and collected data from some
65 individual test sites (Figure 1).
2. MANAGEMENT
The deployment preparations were directed by a management team
comprising representatives of the CSIRO Office of Space Science Applications
(COSSA); CSIRO Division of Exploration and mining; the University of New
South Wales and the Australian Mining Industry Research Association, with
NASA HQ and COSSA acting as the signatories for the mission.
In April 1993 a five-day Radar Image Processing and Applications
workshop was held in Sydney for the participating investigators. In addition to
presenting a theoretical background to the processing of multi-polarised data sets,
the workshop sought to outline SAR calibration and ground sampling procedures;evaluate current applications of SAR in geology, vegetation, soils and soil moisture
and sea-state investigations; examine the interferometric mode of SAR for surface
mapping and to provide participants with hands on experience in basic image
processing of radar data. Guest speakers and workshop leaders included Craig
Dobson from the University of Michigan, Anthony Freeman from JPL and Fred
Kruse from the University of Colorado.
37
Duringthedeploymentdatawasacquiredfor:
i) NASA/Australiancollaborativeprojects;
ii) SIR-Ccalibrationinvestigations;
iii) specificCSIRO-basedresearchprograms;and
iv) aseriesof individualinvestigationsforgovernmentagenciesandprivatesectorsponsors.
Towardstheendof 1994anevaluationworkshopwill beheldtodiscusstheresultsof themissionandallowindividualinvestigatorstopresenttheirfindings.
3. SCIENCE OBJECTIVES
Radar remote sensing technology is comparatively untried, unresearched
and unproven in Australian terrains. SIR-A and SIR-B data in the early 1980's
did provide limited opportunities to investigate and map selected geological andvegetational patterns (Richards et al. 1987).
One of the major objectives of this deployment is to determine thecontribution of AIRSAR and TOPSAR datasets to landform determination and
structural mapping in regolith dominated terrains. One CSIRO research project
sponsored by mineral exploration companies has the following aims:
i) To differentiate surficial regolith materials on the basis of their surface
roughness and dielectric characteristics, especially weathered rock outcrop,lag gravels, soils and vegetation in both mafic and felsic terrains, and to
quantitatively analyse the radar frequency information and the polarimetric
signatures that describe each land component. Information describing
these surface variables should allow recognition of the three fundamentalregimes of a weathered landscape, that is residual, erosional and
depositional. It is anticipated that the processed radar data will provide
significant information concerning the degree of weathering, wind and/or
water erosion processes, and on interpretation of patterns of sedimcntation
and relative disposition within the dispositional units;
ii)
iii)
To display subtle geomorphological features (micro-relief) involving relief
escarpments, drainage and various landforms which may be surface
indicators of subsurface geological structures of exploration significance;
To investigate the capability of polarimetric radar data to map sub-surface
geometries and subtle hidden structures in areas of thinly-covered, gcntlydipping strata;
iv) To determine the optimum viewing and imaging parameters for future use
of satellite and airborne radars for regolith-landform and geologicalmapping in the Australian semi-arid and arid zones; and
v) To generate high resolution digital elevation models of the study areasusing TOPSAR radar interferometry. The models will be used to
geometrically rectify the AIRSAR polarimetric and ERS-1 SAR data
respectively and to assist definition of landform regimes, regolith
characteristics, sources of materials and local regolith stratigraphy. These
topographic datasets will be registered to other remotely-sensed,geological, geophysical and geochemical datasets and used for fault
38
mappingandidentification,terrainanalysisandterrainprocessesanalysis,andestablishinggeochemicaldispersionprocessesandpatterns.
Anothermajorobjectiveincludesinvestigatingtheuseof radarforvegetationmappingin forests,woodlandsandrangelandenvironmentsandfortestingmodelsdevelopedtoaccountforthefull interactionof backscatterfromdifferenttreemorphologiesoverdiffuseground.A numberof investigationsarecentredin theNorthernTerritoryextendingalongatransectfromnearDarwininthenorthtoKatherinein thesouth.Thistransectprovidesaclimaticallydeterminedgradientwhichbringsatransitionfromwetlands,tropicalforestsandwoodlands,savannagrasslandtosemi-aridanddeserts.Theaccuratediscriminationof thesebiomesandtheirboundaryeffectsareseenascrucialtothespatialmodellingof ecosystemsatbothalocalandaglobalscale.
Landdegradationprocessesassociatedwithsalinityandalteredgroundwaterconditionswill bestudiedinanumberof sitesthroughoutAustralia.Onesite,KeranginCentralVictoria,willbeusedasamajorcalibrationsitefortheforthcomingSIR-Cmissionaswellasforhydrogeology.
A jointNASA/Australiaprojectin theGreatSandyDesertregionofWesternAustraliawill usebothAIRSARandTOPSARdatafordetailedreconstructionof Australia'spalaeoclimateandpalaeohydrologyduringtheLateQuaternaryperiod.It isanticipatedthatthesedatasetswill assistin:
* mapping ancient shoreline ridges defining the extent and height of former
lakes;
* mapping lacustrine units and distinct flood sedimentation units such as
slackwater deposits; and
* identifying levee systems of prior streams and the presence of strandline
deposits within dunal corridors.
TOPSAR will be crucial for determining local slope, reconstructing the drainage
network and modelling flood estimation and extent, surfrace run-off and landform
development.
4. CONCLUSIONS
The Australia deployment has provided a core group of Australian researchers an
exciting opportunity to exploit the unique capabilities of both AIRSAR andTOPSAR datasets. The benefits of radar remote sensing technology to earth
system sciences now depends on the regular availability of these precision datasets
from operational spaceborne systems.
5. REFERENCES
Richards, J.A., B.C. Forster, A.K. Milne, J.C. Trinder, and G.R. Taylor, 1987,
"Australian multi-experimental assessment of SIR-B (AMAS). Final Report to
NASA and JPL, March 1987," pp. 11 plus appendices.
39
N95- 23949
CURRENT AND FUTURE USE OF TOPSAR DIGITAL TOPOGRAPHIC DATA FORVOLCANOLOGICAL RESEARCH
Peter J. Mouginis-Mark, Scott K. Rowland and Harold Garbeil
Planetary GeosciencesDepartment Geology and Geophysics, SOEST
University of HawaiiHonolulu, Hawaii 96822
Introduction
In several investigations of volcanoes, high quality digital elevation models (DEMs) arerequired to study either the geometry of the volcano or to investigate temporal changes in reliefdue to eruptions. Examples include the analysis of volume changes of a volcanic dome (Fink etal., 1990), the prediction of flow paths for pyroclastic flows (Malin and Sheridan, 1982), and thequantitative investigation of the geometry of valleys carved by volcanic mudflows (Rodolfo andArguden, 1991). Additionally, to provide input data for models of lava flow emplacement,accurate measurements are needed of the thickness of lava flows as a function of distance from
the vent and local slope (Fink and Zimbelman, 1986). Visualization of volcano morphology isalso aided by the ability to view a DEM from oblique perspectives (Duffield et al., 1993).
Until recently, the generation of these DEMs has required either high resolution stereo airphotographs or extensive field surveying using the Global Positioning System (GPS) and otherfield techniques. Through the use of data collected by the NASA/JPL TOPSAR system, it isnow possible to remotely measure the topography of volcanoes using airborne radarinterferometry (Zebker et al., 1992). TOPSAR data can be collected day or night under anyweather conditions, thereby avoiding the problems associated with the derivation of DEMs fromair photographs that may often contain clouds. Here we describe some of our initial work onvolcanoes using TOPSAR data for Mt. Hekla (Iceland) and Vesuvius (Italy). We also outlinevarious TOPSAR topographic studies of volcanoes in the Galapagos and Hawaii that will beconducted in the near future, describe how TOPSAR complements the volcanology
investigations to be conducted with orbital radars (SIR-C/X-SAR, JERS-1 and ERS-1), and placethese studies into the broader context of NASA's Global Change Program.
TOPSARThe TOPSAR instrument is a C-band (5.6 cm wavelength) radar flown on board a NASA
DC-8 aircraft (Zebker et al., 1992; Evans et al., 1992). Topographic data collected by TOPSARhave a spatial resolution of 5 to 10 m, with a vertical accuracy of 1 to 5 m depending upon therelief of the target -- smoother surfaces (i.e., at the pixel-scale, those surfaces that have auniform relief) will have lower height errors than mountainous areas because of the greateruncertainty in characterizing an inhomogeneous pixel with a single height value. TOPSARswaths are 30 km x 6.4 km in size. These topographic data are acquired concurrently with radar
backscatter images at C-band, L-band (24 cm) and P-band (68 cm), which enable surface texturesand structure to be investigated in a manner comparable to conventional radar analyses ofvolcanoes (e.g., Gaddis et al., 1989, 1990; Campbell et al., 1989).
TOPSAR measurements differ significantly from DEMs derived from the interpolation ofdigitized contour maps. In TOPSAR data sets, a height measurement is made at each pixel andtherefore the TOPSAR DEM provides a truer representation of the surface relief. We note,however, that in its current configuration TOPSAR does not have any absolute geodetic control,so that each TOPSAR scene must be referenced to a geodetic grid before absolute elevations andregional slopes (i.e., those slopes measured along or across the entire swath) can be resolved.
41
Mt. Hekla and Vesuvius
We illustrate the use of TOPSAR data for volcanological research with examples derivedfrom data collected over Mt. Hekla (Iceland) and Vesuvius (Italy) in the summer of 1991.
For Hekla volcano, the thickness of lava flows can be measured from TOPSAR data (Fig.1), and these thickness measurements can be used with existing numerical models of lavarheology to infer yield strengths of 5000 to 30,000 Pa. (Rowland et al., 1992), comparable tolavas of similar composition (basaltic andesite) elsewhere. In all cases, the calculated yieldstrength of a flow increases with distance from the vent (reflecting the greater amount of coolingof the furthest-traveled lava), although average flow thickness remains fairly constant. TOPSARdata also permit the geometry of a moberg ridge (which is a volcanic feature formed by a sub-glacial eruption) to be determined (Fig. 2). Although only a few examples of moberg ridgesaround Hekla were imaged by TOPSAR, it would be possible to use TOPSAR to measure thevolume of many ridges, thereby enabling the amount of lava erupted from different vents to bedetermined (Evans et al., 1992).
TOPSAR data for Vesuvius, Italy,permit the geometry of the volcanoflanks to be determined (Mouginis-Markand Garbeil, 1993). There is a largenumber of valleys on the flanks of theolder portion (Mt. Somma) of Vesuvius.These valleys are primarily water-carvedin origin, but they have also been used aspathways for pyroclastic flows.TOPSAR enables a slope map of theflanks to be derived, and valleygeometry to be measured. Slope mapscan provide valuable input whenexamining the likelihood that differentareas will be affected by volcanichazards such as pyroclastic flows, lavaflows, and lahars. We have found the
slopes of the flanks of Mt. Somma to betypically -13 - 24 ° at lower elevations,-25 - 36 ° closer to the rim, with amaximum value of > 48 ° on the innerwalls of the craters.
sso ALO_
__k_ CROSS FLOW
F440 .... , .... , .... , .... , .... , ....
S00 1000 1500 2000 2500 3000
DISTANCE (m)
Fig. 1: TOPSAR profiles across and along asingle lava flow erupted from Mt. Hekla, Iceland.
51111 ¸
E 52o-
I.-
(_ 500-mW
800
750
E to0
I-- 650
L'3600
tM"!"
550
50O
0
1 •
800
75O
E.._ 700
k-650
LM 600"I"
550
500.... I .... ! .... • - • - i .... i ....
500 1000 1500 500 1000 1500
DISTANCE (m) DISTANCE (m)
800
750
7OOA
E"" 650
I"
'I- 6O0
14.1"r" 550
500
3,
500 1000 1500
DISTANCE (m)
Fig. 2: 3 profiles across a single moberg ridge close to Mt. Hekla, Iceland, derived fromTOPSAR data. Length of moberg ridge perpendicular to these profiles is -3800 m.
42
Profiles down the length of individuals valleys (Fig. 3a) can also be determined.
Potentially, it may be possible to use TOPSAR data to recognize different lithologic units (eitherdifferent in the mechanical properties or absolute age) by this method, since the degree of erosion
of materials on a given slope should vary as a function of strength and/or age. In order to explorethe physical basis for these relationships, a more rigorous study would be required to identify andevaluate the possible significant variables (e.g., Knighton, 1974). The profiles across a valley atdifferent distances from the rim of Mt. Somma (Fig. 3b) also show that the geometry of the
valley varies from one place to another. From our morphologic data, there appears to be a trendtowards proportionally deeper valleys with increasing slope. In general, for each valley, onslopes > 30 ° depth can be twice that on slopes < 10 °. This may be interpreted to be an indicationof the lower erosive action of the flows (either debris flows or running water) on the lower slopes
than on steep slopes, or of deposition on the lower slopes.
800"
600
400"
200
0
-200500 1000 1500
DISTANCE (m)
2000
Fig. 3a (Left, at top): Profiles forindividual valleys on the flanks of Mt.Somma. Fig. 3b (left, at bottom)Cross-sections through a single valleyon flanks of Mt. Somma, as
determined from TOPSAR data. Only
a few representative cross-sections areshown, but the original data wereobtained at every 250 m down theflanks.
40 ¸
20
I--
0 0
UJ
LU> -20I
I-<(...iuJ -40n.-
-6O
R_ILE #1
0 25 50 75 100 125 150 175 200
DISTANCE ALONG PROFILE (m)
Future TOPSAR data sets
To date, only a few TOPSAR datasets have been collected over
volcanoes; the most useful are the datafor the western Galapagos Islands.These volcanoes constitute one of the
Space Shuttle Radar (SIR-C/X-SAR)"Super-Sites", and are also one of thetargets for the JERS-1 VerificationProgram. Use of the TOPSAR DEM,as well as the AIRSAR multi-
wavelength backscatter images and theJERS-1 SAR data, will therefore aid in
the analysis of these orbital data sets.
Several structural and volcanological investigations of the Galapagos volcanoes have been
conducted using air photographs (Chadwick and Howard, 1991) and satellite images (Munro and
Mouginis-Mark, 1990; Rowland and Munro, 1992), but the detailed topography of the islands ispoorly known. Fernandina and Isabela Islands were imaged in May 1993 by TOPSAR, but at thetime of writing the analysis of these data has not been initiated. The use of TOPSAR data to
investigate the spatial distribution of rift zones through the generation of slope maps, themeasurement of lava flow thickness on different slopes, and the calculation of volumes of cindercones and summit calderas, should all significantly improve our knowledge of these infrequentlyvisited volcanoes.
Two TOPSAR deployments over Kilauea Volcano, Hawaii, are planned for September andOctober 1993. These flights will enable the derivation of a "topographic difference map", which
43
shouldpermit the quantitativemeasurementof the volume of new lava eruptedduring theinterveningmonth. This will allow anaverageeffusionratefor thevolcanoto bedetermined,aswell asfacilitate the studyof the growthof acompoundlava flow field. Finally, aspart of theDecadesVolcano program (Bennett et al., 1992), planning is under way for a TOPSARdeploymentto SantaMaria Volcano,Guatemala,aswell asthecollectionof ERS-1orbitalSARdatafrom the temporarygroundreceivingstationin Atlanta,USA. Sometimein thefuture wehope to use TOPSAR to aid the analysisof the Santiaguitolava dome, and help with theproductionof newhazardmapsthroughtheconstructionof detailedtopographicmaps.
Acknowledgments
This research was supported by NASA grants NAGW-1162 and NAGW-2468 fromNASA's Office of Mission to Planet Earth.
References
Bennett EHS, Rose WI, Conway FM (1992) Santa Maria, Guatemala: A Decade Volcano. Eos73:521 - 522
Campbell BA, Zisk SH, Mouginis-Mark PJ (1989) A quad-pol radar scattering model for use inremote sensing of lava flow morphology. Remote Sens Environ 30:227 - 237
Chadwick WW, Howard KA (1991) The pattern of circumferential and radial eruptive fissureson the volcanoes of Fernandina and Isabela islands, Galapagos. Bull Volcanol 53:257 - 275
Duffield W, Heiken G, Foley D, McEwen A (1993) Oblique synoptic images, produced fromdigital data, display strong evidence of a "new" caldera in SW Guatemala. J VolcanolGeotherm Res 55:217 - 224.
Evans DL, Farr TG, Zebker HA, van Zyl JJ, Mouginis-Mark PJ (1992) Radar interferometricstudies of the Earth's topography. Eos 73:553 and 557 - 558
Fink JH, Malin MC, Anderson SW (1990) Intrusive and extrusive growth of the Mt. St. Helenslava dome. Nature 348:435 - 437
Fink JI-I, Zimbelman JR (1986) Rheology of the 1983 Royal Gardens basalt flows, KilaueaVolcano, Hawaii. Bull Volcanol 48:87 - 96
Gaddis L, Mouginis-Mark P, Singer R, Kaupp V (1989) Geologic analyses of Shuttle ImagingRadar (SIR-B) data of Kilauea Volcano, Hawaii. Geol Soc Amer Bull 101:317 - 332
Gaddis LR, Mouginis-Mark PJ, Hayashi JN (1990) Lava flow surface textures: SIR-B radar
image texture, field observations, and terrain measurements. Photogram Eng Remote Sensing56:211 - 224
Knighton AD (1974) Variation in width-discharge relation and some implications for hydraulicgeometry Geol Soc Amer Bull 85:1069 - 1076
Malin MC, Sheridan MF (1982) Computer-assisted mapping of pyroclastic flows. Science 217:637 - 640
Mouginis-Mark, PJ and Garbeil, H. (1993) Digital topography of volcanoes from radar
interferometry: An example from Mt. Vesuvius, Italy. Submitted to Bulletin Volcanology.Munro DC, Mouginis-Mark PJ (1990) Eruptive patterns and structure of Isla Fernandina,
Galapagos Islands from SPOT-1 HRV and Large Format Camera images. Int J RemoteSensing 11:15011 - 1174
Rodolfo KS, Arguden AT (1991) Rain-lahar generation and sediment-delivery systems atMayon Volcano, Philippines. In: Sedimentation in Volcanic Settings, SEPM Sp. Pub. No. 45:71 - 87
Rowland SK, Munro DC (1992) The caldera of Volcan Fernandina: a remote sensing study of itsstructure and recent activity. Bull Volcanol 55:97 - 109
Rowland, SK, Garbeil, H. and Mouginis-Mark, PJ (1992). Yield strengths of recent Hekla lavas.Fall 1992 AGU Abstracts, p. 648.
Zebker HA, Madsen SN, Martin J, Wheeler KB, Miller T, Lou Y, Alberti G, VetreUa S, Cucci A
(1992) The TOPSAR interferometric radar topographic mapping instrument. IEEE TransGeosci Rem Sen 30:933 - 940.
44
N95- 23950
PRELIMINARY ANALYSIS OF THE SENSITIVITY OF AIRSAR
IMAGES TO SOIL MOISTURE VARIATIONS
Rajan Pardipuram, William L. Teng
Hughes STX Corporation, Lanham, MD 20706
James R. Wang, Edwin T. Engman
NASA, Goddard Space Flight Center,
Greenbelt, MD 20771
Introduction
Synthetic Aperture Radar (SAR) images
acquired from various sources such as Shuttle
Imaging Radar B (SIR-B) and airborne SAR (AIRSAR)
have been analyzed for signatures of soil moisture
(Dobson et al., 1986, Wang et al., 1986, Rao et
al., 1992). The SIR-B measurements have shown a
strong correlation between measurements of surface
soil moisture (0-5 cm) and the radar
backscattering coefficient ,o, (Wang et al.,
1986). The AIRSAR measurements, however, indicated
a lower sensitivity (Rao et al., 1992). In this
study, an attempt has been made to investigate the
causes for this reduced sensitivity.
Measurements
Polarimetric AIRSAR data were acquired over
the Little Washita watershed near Chickasha,
Oklahoma during June 10-18, 1992. A total of 8
days of flights were made during this period.
There was a series of heavy rainfall prior to June
i0. No rainfall was reported between June i0 and
18. Soil moisture samples in the top 5 cm layer
were collected at a number of fields during the
time of the flights. The average soil moisture was
-0.26 gm/cm 3 on the first day of flight (June I0)
and "0.13 gm/cm 3 on the last day of flight (June
18).
Two areas covered by the AIRSAR flights were
selected for the study, one southwest of the
watershed (site i), and the other northeast of the
watershed (site 2). Three sets of images (C, L,
and P-bands) for the two areas, acquired on three
different dates, June i0, 14, and 18, were
analyzed. In order to obtain a broader perspective
on the sensitivity of the SAR images to soil
moisture variations, a finite strip of 200 pixels
in the cross track and 1024 pixels in the along
track directions were chosen from each image. The
45
strips from each scene were chosen such that they
cover approximately the same area on the ground.
Results
The results from the analysis for site 1 are
shown in Figures 1 and 2. Each data point in these
figures represents an average of 200 pixels in the
cross track and 8 pixels in the along track
directions. Averaging was performed to reduce the
effect of speckle and noise.
Figures 1 and 2 indicate the variations of
ahh O, at all three frequencies, for the June I0aH_ 18 images. These figures show that (i) the
average value of ahh ° changed only by about 1 dB,2.5 dB, and 3 dB, _or C, L, and P-bands,
respectively from June i0 to 18. whereas soil
moisture changed by ~0.13 gm/cm 3 during the same
period; and (2) amplitude variations within the
strips are much higher in comparison (on the order
of 5-8 dB). Since soil moisture is not expected to
differ by a significant amount within a strip, thewide amplitude fluctuations indicate that the
radar backscatter of the AIRSAR images is
sensitive to other surface features. The general
pattern of the amplitude variations of ahh ° is thesame for both the June I0 and 18 images, which
suggests that these variations are caused by
surface features which did not change from June i0
to 18. However, at this point, it is not clear
which surface feature/features are causing these
variations. Comparison of responses of the three
frequencies shows that P-band has the highest
variation (standard deviation of " 2.2) and C-band
the lowest (standard deviation of 0.6).
Images of site 2 were also analyzed to
determine if they indicate similar trends.
However, a disturbing feature was observed in the
C-band images. Figures 3 and 4 indicate the
variations of a ° values for C-band, for June i0
and 18, respectively. These figures show that,
while ahh ° is higher than avv ° on June I0, avv ° ishigher _Han ahh ° on June 18. This pattern was not
noticed in the case of L and P-bands. This feature
is probably caused by an error in the calibration
procedure; therefore, these images were not used
for the analysis.
Conclusions
An attempt was made to examine the causes for
the lower sensitivity of AIRSAR images to soil
46
moisture variations in comparison with that ofSIR-B images. Based on the results obtained, itcan be inferred that a° values are less sensitiveto soil moisture than to other surface features.Further analysis of these images is required toidentify those surface features which predominatethe radar backscatter in the case of AIRSAR.
Some of the C-band images indicated a changein the dominant polarization with time. Thischange is not expected to occur over a typicalagricultural area and could be due to a potentialproblem in the calibration.
References
Dobson, M. C. and F. T. Ulaby, "PreliminaryEvaluation of the SIR-B Response to Soil Moisture,Surface Roughness, and Crop Canopy Cover," IEEETrans. Geosci. and Remote Sensinq, GE-24(4). DD.
517-526, 1986.
Rao, K. S., W. L. Teng, and J. R. Wang, "Terrain
Effects on Backscattering Coefficients Derived
from Multifrequency Polarimetric SAR Data,"
IGARSS'92, vol. i, pp 86-88, Houston, Texas, 1992.
Wang, J. R., E. T. Engman, J. C. Shiue, M. Rusek
and C. Steinmeier, "The SIR-B Observations of
Microwave Backscatter Dependence on Soil Moisture,
Surface Roughness, and Vegetation Covers," IEEE
Trans. Geosci. and Remote Sensing, GE-24(4), pp.
510-516, 1986.
47
c O
0.0June 10, HH Pol., look angle -40 °
.j'.,":'", ..'-.'. k .". . : . .-I0.0 ."'v.. ,.: " .,..... 'J'..: ", :,, :V-.". ,,;'% -
i : , . . "t:
I"% ,/"i _/ i ./'_"'_ _ ; h.7". ;I",I \ r\ t _
-20.0 '; ,.;i
-- C-band........ L-band
........P-band-30.0
ave. of 200 pixetscross track and 8 along track
I I , I
"40"00.0 40.0 80.0 120.0
Pixel number
Figure i. The along track variation
of O_HO at the look anqle of 40 ° ,
for c, L, and P-bands (June i0,1992) .
0.0
-10.0
-20.0
-30.0
June 18, HH Pol., look angle-53 °
/,.-,] ,.: : ,., ,. , "]
I" i_'C, .j "::;,:;/ , :': '. .: : ,: ,.
;., .. ::: :, ._ -' .., ..: ",,: :" ..:
, , , , ... ,, .'.:" ,,'- ,,-" '.....'...'lJ
,; "J _ i' / t
f-- C-band
........ t_-band
........ P-band
ave. of 200 pixels cross track and 8 along track
. I I I
-40"00.0 40.0 80.0 120.0
Pixel number
Figure 2. The along track variation
of oHH ° at the look angle of 53 ° ,
for _ L, and P-bands (June 18,1992) .
v
oO
0.0June 10, C-band, look angle -37 °
t i i
-10.0
-20.0
-30.0
"c
-- HH
VV
HV
ave of 200 p_xels cross track and 8 along track
-40.0 _ -0.0 40.0 80.0 120.0
P_×PI number
Figure ._. ]'he along track variation
of C-ban_ o ° values, at the lookana}e cf ; c_ 5or }{H VV and HA7, i t
po]ar1::4t !ons (,June ]0, 1992).
o
0.0June 18, C-band, look angle-34 °
i i i
-I0.0
-20.0
/" " . / " . .'., "." ,
-- HH- VV
-30.0 [- HVave of 200 pixels cross track and 8 along track
-40.0 _ J J_.0.0 40.0 80.0 120.0
Pixelnumber
Figure 4. The along track variation
of C-band o ° values at the look
angle of 34 °, for HH, W, and HV
polarizations (June 18, 1992).
48
N95- 23951
UNUSUAL RADAR ECHOES FROM THE GREENLAND ICE SHEET
E. J. Rignot, J. J. van Zyl, S. J. Ostro, and K. C. Jezekt
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109
tByrd Polar Research Center, Ohio State University, Columbus, OII 43210
1. INTRODUCTION
In June 1991, the NASA/Jet Propulsion Laboratory airborne synthetic-aperture radar
(AIRSAR) instrument collected the first calibrated data set of multifrequency, polarimetric,
radar observations of the Greenland ice sheet (Rignot et al., 1993). At the time of the AIRSAR
overflight, ground teams recorded the snow and firn (old snow) stratigraphy, grain size, density,
and temperature (Jezek and Gogineni, 1992) at ice camps ill three of tile four snow zones iden-
tified by glaciologists to characterize four different degrees of summer melting of tile Greenland
ice sheet (Benson, 1962). Tile four snow zones are: 1) the dry-snow zone, at high elevation,
where melting rarely occurs; 2) the percolation zone, where summer melting generates water
that percolates down through the cold, porous, dry snow and then refreezes in place to form
massive layers and pipes of solid ice; 3) the soaked-snow zone where melting saturates the snow
with liquid water and forms standing lakes; and 4) the ablation zone, at the lowest elevations,
where melting is vigorous enough to remove the seasonal snow cove," and ablate the glacier ice.
There is interest in mapping the spatial extent and temporal variability of these different snow
zones repeatedly by using remote sensing techniques. The objectives of the 1991 experiment
were to study changes in radar scattering properties across the different melting zones of tile
Greenland ice sheet, and relate the radar properties of the ice sheet to tile snow and firn physical
properties via relevant scattering mechanisms, llere, we present an analysis of the unusual radar
echoes measured from the percolation zone.
2. EXPERIMENTAL RESULTS
Figure 1 shows average values of the radar reflectivity o'_ L (i.e., receiving right-circular
polarized signals and transmitting left-circular polarized signals) obtained by averaging the
radar measurements recorded by AIRSAR at the Crawford Point site in the percolation zone
along the flight path as a function of the incidence angle of the radar illumination O. At 5.6
and 24 cm, a_t L is higher than unity at 18 °, and decreases toward higher incidence angles. At
68 cm, cr°RL is ten times lower, and shows kilometer-scale spatial variations. Figure 1 also shows0 0 0 0
the circular polarization ratio, pc = ffFtR/O'RL, and the linear polarization ratio PL = O'ItV/OHft
obtained at the Crawford Point site. These ratios of echo power in orthogonal senses are defined
to be equal to zero for specular backreflection from a perfectly smooth dielectric surface, pc is
larger than unity at 5.6 and 24 cm for incidence angles larger than 30 ° and 45 °, respectively,
increasing to 1.6 and 1.4 at 66 ° . At 68 cm, #c is everywhere less than 0.8 and drops as low
as 0.1 in some places, with kilometer-scale spatial variations negatively correlated with those
observed in the radar reflectivity images. #L is as large as 0.46 at 5.6 cm and 0.22 at 24 cm,
but remains less than 0.1 at 68 cm.
In the AIRSAR scenes of the Swiss camp and the GISP II sites, at all three wavelengths,
O'_L are 10 to 30 times lower than at Crawford Point; pc is less than 0.4, and PL is less than 0.1.
To the best of our knowledge, no natural terrestrial surface other than the Greenland percolation
zone shows strong echoes with Pc > 1 and #L > 0.3 (see Figure 1 caption). However, strong
echoes with large values of tLc and #L have been reported for the icy Galilean satellites since the
49
1970's(Ostroet al., 1992).More recently,radarobservationsof the Marsresidualsouthpolaricecap(Muhlemanet al., 1991),portionsof Titan (Muhlemanet al., 1990),andpolar capsonMercury(Sladeet al., 1992)haverevealedthat surfaceswith highradarreflectivityand#c > 1existelsewherein thesolarsystem.Figure1showsthat themeanvaluesof disk-integratedradarreflectivitiesa_) c and circular and linear polarization ratios pc and/2 L for Europa, Ganymede,
and Callisto at 3.5 and 13 cm resemble those of the percolation zone at 5.6 and 24 cm, and dwarf
the values reported for other terrestrial surfaces. Yet, Greenland's average values at 24 cm are
several tens of a percent lower than at 5.6 cm, and tLc < 1 at 68 cm, indicating a change in the
scattering process at the longer wavelengths, whereas 70-cm estimates of tLc for the icy satellites
apparently exceed unity (Campbell et al., unpublished data). Also, #c for the percolation zone
decreases significantly from 66 ° to 18 °, whereas no such difference has been noticed for the icy
satellites (Ostro et al., 1992); and O'_L is a nmch stronger function of the incidence angle than
in the case of the icy satellites (Ostro et al., 1992).
3. INTERPRETATION
Zwally (1977) suggested that ice inclusions could exl)lain low emissivities measured for the
percolation zone by spaceborne microwave radiometers. Since then, surface-based radio sounding
experiments, and airborne active and passive microwave nmasurements (Swift et al., 1985), have
supported the hypothesis that volume scattering from subsurface ice layers and ice pipes is the
major influence on the radar returns. Recent surface-based radar observations conducted at
Crawford Point (Jezek and Gogineni, 1992) at 5.4 and 2.2 cm provided clear evidence that,
at incidence angles between 10 ° and 70 ° , most of the scattering takes place in the most recent
annual layer of buried ice bodies. Studies of the snow stratigraphy at Crawford Point at the time
of the radar flight indicate that the ice inclusions from the previous summer melt were at 1.8 m
below the surface. Ice layers and ice lenses, a millimeter to a few centimeters thick, extend at
least several tens of centimeters across, parallel to the firn strata. Ice pipes, several centimeters
thick and several tens of centimeters long, are vertically extended masses reminiscent of the
percolation channels that conduct meltwater down through the snow during summer, feeding
ice layers. The fact that radar returns measured at 68 cm are significantly weaker and have lower
polarization ratios than those at 5.6 and 24 cm suggests that the discrete scatterers rest)onsible
for the radar echoes are of typical dimension less than a few tens of centimeters, similar to the
scales of the solid-ice inclusions. The 68-cm echoes probably are dominated by single reflections
from deeply buried layers of denser firn or concentrated ice bodies, whereas the 5.6- and 24-cm
echoes probably are dominated by multiple scattering from the ice layers and pipes in the most
recent annual layer. The relatively sharp decrease in Pc and #c for 0 less than 40 ° perhaps
reveals the presence of a strong, specular reflection from the ice layers at small incidence angles,
which is also suggested by the strong dependence of radar reflectivity on incidence angle. Ice
layers and pipes also form in the soaked zone, but the snow there is so saturated with liquid water
that the radar signals are strongly attenuated, cannot interact with the buried ice formations,
and hence yield echoes with low reflectivities and polarization ratios. In the dry-snow zone, the
snow is dry, cold, porous, clean, and therefore very transparent at microwave fl'equencies, but
does not contain solid-ice scatterers that could interact with the radar signals.
For the satellites, no in-situ measurements exist, but theoretical interpretations favor sub-
surface coherent volume scattering as the source of the radar signatures (Hapke, 1990), a phe-
nomenon also known as weak localization (see van Albada et al., 1990 for a review). Coherent
backscattering can theoretically produce strong echoes with pc > 1 (the helicity of the incident
polarization is preserved through multiple forward scattering) and #L _ 0.5, provided that (i)
5O
the scatteringheterogeneitiesarecomparableto or larger than tile wavelength(Peters,1992),and (ii) the relativerefractiveindexof the discrete,wavelength-sizedscatterersis smallerthan1.6(Mishchenko,1992). As notedby Ostroand Shoemaker(1990),prolongedimpact crater-ing of the satellitesprobablyhasled to the developmentof regolithssimilar in structureandparticle-sizedistribution to the lunar regolith,but thehigh radartransparencyof icecoml)aredwith that of silicatespermitslongerphotonpath lengths,and higher-orderscattering, tlencecoherentbackscattercandominatethe echoesfrom Europa,Ganymede,and Callisto, but con-
tributes negligibly to hlnar echoes. Similarly, the upper few meters of the Greenland percolation
zone are relatively transparent (unlike the soaked zone) and, unlike the dry-snow zone., contain
an abundance of solid-ice scatterers at least as large as the radar wavelength, with a relative
refractive index of about 1.3, so coherent backscatter also can dominate the echoes there, itow-
ever, the detailed subsurface configurations of the satellite regoliths, where heterogeneities are
the product of meteoroid bombardment, are unlikely to resemble that within the Greenland
percolation zone, where heterogeneities are the 1)roduct of seasonal inciting and freezing.
Acknowledgements: We thank Dr. R. Thomas, llead of the Polar Research Program at
NASA Ileadquarters, for supporting this research. Part of this work was carried out at the
Jet Propulsion Laboratory, California Inslitule of Technology, under contract wilh the National
Aeronautics and Sl)ace Adminislration.
REFERENCES
C. S. Benson, US Army Snow Ice 1)ermafi'ost Iles. Estab. Res. Rep. 70 (llanover, Nit, 1962).
K. C. Jezek and S. P. Gogineni, IEEE Geosc. Rcm..5'_ts. £'oc. Newsletter 85, (i (1992).
B. Ilapke, Icarus 88,407 (1990).
M. I. Mishchenko, Earth, Moon, and Planets 58, 127 (1992).
D. O. Muhleman, B. J. Butler, A. W. Grossman, and M. A. Slade, Scic_zcc 253, 1508 (1991).
D. O. Muhleman, A. W. Grossman, B. J. Butler, and M. A. Slade, ,9cic_wc 248,975 (1990).
S. J. Ostro el al., J. Gcophys. Rcs. 97, 18277 (1992).
S. J. Ostro and E. M. Shoemaker, Icarus 85,335 (1990).
K. J. Peters, Phys. Rev. B 46, ,_01 (1992).
E. Rignot, S. Ostro, J. van Zyl, and K. Jezek, Science 261, 1710 (1993).
M. A. Slade, B. J. Butler, and I). O. Muhleman, ,5'cicnc_: 258,635 (1992).
C. T. Swift, P. S. llays, J. S. llerd, W. L. Jones, and V. E. Dehnore, J. Geophys. Res. 90, 1983
(1985).
M. P. van All)ada, M. B. van der Mark and A. Lagendijk, in Scattering a_M Localization of
Classical Waves in Random Media, P. Sheng, Ed. (World Scientific, Singapore, 1990), p. 97.
II. J. Zwally, J. Glaciol. 18, 195 (1977).
51
3.0
>, 2.5.i
>ol
"o 2.0
1.5
_ 1.0_oo
0.0
2.00
1.5
i.o
•_ 0.5
0.0
0.60
.I
co 0.4
N"= 0.3
0
'' 0.2
._c 0.1--I
0.(
.... I .... I .... I .... I .... I ....
A
G
lC
0 20 30 40 50 60 70
Incidenceangle(degrees),
'''l .... l .... I .... I''''I ....
0 20 30 40 50 60 70
Incidenceangle(degrees).... I''''t''''1''''1''''1''''
C
-EGC
Oc
, , q , l _ , L J I , , ' I J I i I I i I I I I I I I I i
0 20 30 40 50 60 70
Incidence angle (degrees)
Figure 1. Average values of the (A) OC radar reflectivity O'_L , (B) circular polarization0 0 0 0ratio Pc = aRR/aRL, and (C) linear polarization ratio #L = crHV/CrHtt for the Greenland per-
colation zone obtained by averaging the radar measurements recorded by AIRSAR at Crawford
Point along the flight path, at 5.6 (C-), 24 (L-), and 68 cm wavelength (P-), as a function of
the incidence angle of the radar illumination. Radar backscatter values and polarization ratios
measured from wavelength-sized roughness on a surface of lava (black squares) (J. J. van Zyl, C.
F. Burnette, and T. G. Farr, Geophys. Res. Lett. 18, 1787 (1991)) and from tropical rain forest
(black dots) (A. Freeman, S. Durden, and R. Zimmerman, Proc. of the Int. Geos. Rem. Sens.
Symp., Houston, Texas, May 26-29, IEEE New York Pub., 1686 (1992)) are also shown in the
Figure. The values indicated for the icy satellites Europa (E), Ganymede (G), and Callisto (C)
are disk-integrated average measurements at 3.5 and 13 cm wavelengths (Ostro et al., 1992).
52
B95-23952
MAC EUROPE '91 CAMPAIGN: AIRSAR/AVIRIS DATA INTEGRATION FOR
AGRICULTURAL TEST SITE CLASSIFICATION
S Sangiovanni*, M. F. Buongiorno**, M. Ferrariv, i*** and A. Fiumara*
*Earth Observation Division, Telespazio SpA
Via Tiburtina 965, Rome - Italy
**formerly at Telespazio SpA, Via Tiburtina 965, Rome - Italynow at Istituto Nazionale di Geofisica, Via di Vigna Murata 605, Rome - Italy
***Electronic Engineering Dpt. - Universit_ "La Sapienza"Via Eudossiana 18, Rome - Italy
1. INTRODUCTION
During summer 1991, multi-sensor data were acquired over the Italian test site
"Oltrepb Pavese", an agricultural flat area in Northern Italy. This area has been theTelespazio pilot test site for experimental activities related to agriculture applications.
The aim of the investigation described in the following paper is to assess theamount of information contained in the AIRSAR and AVIRIS data, and to evaluate
classification results obtained from each sensor data separately and from the combined
dataset. All classifications are examined by means of the resulting confusion matrices and
Khat coefficients (Congalton et al., 1983). Improvements of the classification results
obtained by using the integrated dataset are finally evaluated.
2. DATA SET DESCRIPTION
AIRSAR data were acquired with flight Nr. 9 i- 118 on June 22 nd. Four tracks of
the area were available. We used the 45-1 track because it was the closest to the area of
the ground truth acquisition. Among the three available bands, P band was not used
because it was highly disturbed by an interference noise pattern.AVIRIS data were acquired with flight 910719b on July 19 th.
In o_der to obtain info_nation about the ground truth, two "in situ" campaigns
were performed. Due to slight errors in the flight tracks, the images do not exactly match
the field survey area: in fact, about 45% oaly 3fthe SAP, images are covered by ground
truth and this percentage decreases to about 20% in the AVIRIS images.
3. SAR DATA ANALYSIS
3.1. "Per Field" Classification
The discriminant analysis was performed on every combination of the 6 power
"bands" (2 frequencies and 3 polarizations); the conclusions can be summarized: 1) Khat
values range from 0.555 to 0.815 as the number of features per observation is increased;
the optimized choice is the L-VV, C-HH, C-HV combination, which allows a goodclassification accuracy (Khat value is 0.805) with a reasonably small amount of data. 2)
Although the results are quite good, problems still arise in the discrimination of alfalfa andcorn from wheat. These problems were not solved even by adding more "bands". 3) The
Khat values obtained are quite optimistic since the same data set was used both for
training and testing.
3.2 "Per Pixel" Classification
53
In this case we used all features available for the AIRSAR (that is 6 bands in the
NML classifications and 18 for the polarimetric algorithms).
By the examination of the Table !, we can draw the following conclusions: 1) the
classifiers specifically designed for polarimetric data (Lee et al., 1992; Kong et al., 1987)
do not seem to improve the classification accuracy. This could mean that phaseinformation is not vital: in fact, the classification accuracies obtained with the NML
classifier are very close to polarimetric ones, but the first uses less information (only the
power features). 2) Speckle filtering is very useful to improve the classification accuracyas denoted by the 9% gain in the Khat values [(the speckle reduction filters that have been
implemented are adaptive filters (Lee et al., 1991; Frost et al., 1981)].
4. AVIRIS DATA ANALYSIS
4.1. Data Reduction
From the original 224 AVIRIS bands, we selected 131 bet_,een 0.4 pm and 1.75p.m. The radiance to reflectance reduction was performed by means of the "flat field"
technique (Crowley, 1990). For this work we identified a suitable flat field examining
some sand deposits located near the Po river banks. Although this method is not veryreliable, especially in the short wavelengths, it seems to be the best approach in order to
retrieve the relative ground reflectance, since no information on local atmosphericparameters (such as aerosols, water vapor, etc.) were available.
We used the reflectance data obtained to draw spectral signatures of agricultural
crops and other targets present in the area (an example is given in Figure 1). The analysisof the spectra shows a great agreement with experimental ones (Elvidge, 1990; Martin,1990; Goetz, 1991).
The spectral analysis allowed us to select 14 significative bands that maximized
the differences between crops in the reflectance spectra (see Figure !). We obtained a
further reduction by means of a Principal Components Analysis (Loughiin, 1991; Fung,1987) performed on these bands; this analysis allowed us to define two combinations of
components [PC2, PC3, PC6 (referred to as "tell ") and PC I, PC2, PC3 (referred to as
"ref2")]. The PC Analysis was also performed on the 14 radiance bands to investigate the
effects of calibration on the classification accuracy.Other two combinations of components[PC3, PC4, PC5 ("rad 1") and PC 1, PC3, PC4 ("tad2")] were thus provided. In both cases,
the selection was done by visual examination of the single PC images and by the analysisof the eigenvectors' matrix.
4.2. Data Classification
AVIRIS classifications were performed on the same classes as for SAR data and
in a "per pixer' approach. The features used m the analyses are the PC combinations, bothin reflectance and m radiance values, described above.
The Khat values obtained (see Table 2) are close to the ones given by the AIRSAR "per
pixer' classifications and, though not expected, even the "unsupervised results" are verygood. This could be explained as follows: 1) AVIRaS gives very specific information foreach channel: this allows the classifier to recognize different objects without the need of a
preliminary training. 2) In order to reduce the number of generated classes, the clustered
image was post-processed by merging those clusters that seemed to belong to the sameclass.
5. DATA FUSION
5.1. Image Registration
We performed the following operations: 1) slant-to-ground range projection of the6 AIRSAR power speckle-filtered images; 2) re-sampling of AIRSAR images to match theAVIRIS pixel spacing; 3) image-to-image registration with a cubic convolution filter.
54
Steps! and2wereexecutedtogetherwithanewprojectionalgorithmdevelopedbyTelespaziowhichallowedustoobtainlessthanl pixeiinMeanSquaredErrorafterco-registration.
5.2. Multi-Sensor Classification
For comparison purposes, we used the same classes as in the AVIRIS and
A1RSAR "per pixel" classifications. The integrated classification was camed out on the
six projected AIRSAR power images and the AVIRIS PC images. Fifteen combinations
were classified using the NML and clustering algorithmsFrom the Table 3, it is evident that the multi-sensor integration gives good
results: in fact, these values are generally and significantly higher than those achieved by
AIRSAR and AVIRIS separate classifications. The extremely good accuracy obtained by
the integrated data is also demonstrated by the absence of "critical pairs" in the best multi-sensor confusion matrix. This absence can be justified by noticing that each sensor data
classification was not affected by the same "critical pairs", but f'_ both cases it was very
difficult to discriminate alfalfa from other crops (especially from wheat and corn).
6. CONCLUSIONS
In thi:, work a multi-sensor analysis was carried out; first the separate data sets
were processed and classified, then their integration and the consequent classification were
performed.The SAR data analysis showed (see Table i) that polarimetric information does
not seem to improve the classification accuracy: in fact, in spite of the great number of
features available, the polarimetric classifiers show a little improvement with respect to
the NML classifier performed on the power images only (speckle reduction techniques
furtherly improve the classification results).
Among the L and C power information, the co-polarizations (HH, VV) seem to beuseful for classes identification, while the HV polarization turned out to be useful in
"critical pairs" discrimination. It also seems that crops classification is more accurate when
using L band.The great amount of AVIRIS data and its high spectral resolution, which are
surely useful in the characterization of green vegetation, imply two major problems: a)reduction of the data, b) accurate atmospheric corrections. Many efforts are being devoted
to implement simulation models that will allow us to obtain a more accurate atmosphericcorrection.
7. REFERENCES
Congalton, R.G., 1983, "Assessmg Landsat classification accuracy using discrete multivariate analysis
statistical techniques", PE & RS, vol. 49, n" 12, pp. 1671-1678
Crowley, S.K., 1990, "Techniques for AVIRIS data normalization in areas with partial vegetation cover",JPL A VIRIS Workshop, Pasadena (CA).
Eividge, C.D., 1990, "Visible and infrared characteristics of dry plant materials", Int. Journ. of Remote
Sensing, vol 12Frost, V.S., J.A. Stiles, K.S. Shanmugan, J.C. Holtzman, S.A. Smith, 1981, "An adaptive filter for
smoothing noisy radar images", IEEE, vol. 69, n" 1
Fung, T., L.E Drew, 1987, "Application of principal components analysis to change detection",Photogrammetric Engineering and Remote Sensing, vol 53, n" 12, December 1987, pp 1649-1658
Goetz, A.F.H., 1991, "Sedimentary facies analysis using AVIRIS data: a geophysical inverse problem",
JPL A VIRIS Workshop, Pasadena (CA), May 20-21
Kong, J.A., H.A Yueh, A.A. Swartz, 1987, "Identification of terrain cover using the optimum polarimetric
classifier", Journ of E.M. Waves & Appl., vol 2, pp 171-194Lee, J.S., M.R Grtmes, S.A. Mango, 1991, "Speckle reduction in multipolarisation multifrequency SAP,
imagery", IEEE GE 29, pp. 535-544Lee, J.S., M.R. Grunes, 1992, "Feature classification using multi-look polarimetric SAR imagery", Proc.
oflGARSS, pp 77-79
55
Loughlin, W.P., 1991, "Principal component analysis for alteration mapping", Proc. of ERIM "EighthThematic Conference on Geologic Remote Sensing", April 29 - May 2, i 991
Martin, M.E., J.D Abel 1990, "Effects of moisture content and chemical composition on the near infrared
spectra of forest foliage", Proc. ofSPIE, vol 1298, Imaging Spectroscopy of Terrestrial Environment
Table 1. Khat values and 96% Confidence Intervals
for AIRSAR "Per Ptxel" Classifications
Classiflcat_n
NML on LC bands i
NML on LC Lee filtered
NML on LC Frost flltwed
Lee's po_rimet_
Kong's pogarkneUlc
I (H_ T A KHAT
(vqi).8: _ 0.9
q).8, I 0.8
I).8£ ) 0.6
(,8( I 0.8
(,.8_) 1.0
Table 3. Khal values and H% Confidence Intervals
for Multi-Sensor "Per Pixel" Classifications
Table 2. Khat values and 96% Confidence Intervals
for AVIRIS "Per Ptxel" CJal;Mf_JltJofls
Classification
NML on Radl
NML on Refl
NML on Rad2
NML on Rel2
CLUSTER o#! Ref2
CLUSTER on Rail2
CLUSTER on Rel2+Rad2
KHAT A KHAT
0.8108 1.68
0.8261 1.82
0.8318, 1.61
0.851 1.53
:).8408 1.61
0.841 1.58
).8509 1.56
Cla_l_.atton
NML on Mix 1
NML on Mix 2
NML on Mix 3
NML on Mix 4
NML on Refl+PC SAR
NML on Rel2+PC SAR
NML on Refl+LC
NML on Radl+PC SAR
NML on Rad2+PC SAR
NML on Radl+LC
NML on Ref2+LC
NML on Rad2+LC
NML on Mix 8
CLUSTER on Rad2+LC
CLUSTER on R_fl+LC
).
D
),i
3
L!i
I.Ii
i.. (,0.|
0..c
0.[
0._(
0..c
0._
0._
itA A KHAT
32_ 1,62i
88' 1.39
189 L34
Lq( 1.32
)28 1 .I0
[35 1.06
37: 1.05
_t0: 1.02
44_ 0.99
5_ 0.92
58: 0.88
59_ 0.87
51: 0.85
_21 1.67
25_ _.18
Table 4. Additional combinations referred as "Mix-
Y" (Y= I to |) in Table 3
Name
.. Mix-1
Mix.2
Mix.3
F__,___-'es,,-_J_
PC2-i____n,___+ C4tlh ÷ PCI_P..AD.
PC2-Rellec/÷ PC3.p-,4u--___, + PCI-SAR
PC2-Refled + PC3-Radlances + C-HH +
PCI-SAR
PC2.Reflect + PC34RmJiances ÷ C4tH + LJtH ÷PCI-SAR
PC2-Reflect + PC3-Radtances + C-HH + L-HH +
C.HV + PCI-SAR
1OOOO F
14000¸
_ 2000
100O08000.
0 : :
O O)
IIJ j "_ "\
Wavelength (rim)
Figure 1. Typical crops spectral Signatures an(; bands selected for I_A
Sugar be_
........ Tobacco
........... Co_r_
Bands Selec/ed
56
N95- 23953
MEASUREMENT OF OCEAN WAVE SPECTRA
USING POLARIMETRIC AIRSAR DATA
D.L. Schuler
Remote Sensing Division
Naval Research Laboratory
Washington, DC 20375-5351
1.0 BACKGROUND
Azimuthally traveling waves are seldom well imaged in either synthetic-
aperture radar (SAR) or real-aperture (RAR) images of tile ocean (Alpers et al.,
1981). A polarimetric technique has been investigated (Schuler et al., 1993) which
increases a radar's sensitivity to ocean wave tilting when the waves have a component
of their propagation vector in the azimuthal direction. The technique offers
improvement for polarimetric measurements of azimuthal wave slope spectra. A
modification of this technique involving wave-induced changes of the polarization
signature location offers a means of measuring azimuthal wave spectra for both
polarimetric SAR and RAR. This method senses wave-tilts directly and does not
require knowledge of the microwave modulation transler function.
In the case of RAR images, cross section modulation by ocean waves is
normally attributed to two principal sources, tilt modulation and, hydrodynamic
modulation. The effect of these modulations is described mathematically by a complex
modulation transfer function (MTF). For radar images of ocean waves both of these
modulation sources roll-off to zero in the azimuthal direction. Therefore, complete
two-dimensional k-space wave spectra derived from microwave data are often quite
different than the physical ocean spectra. Evidence will be presented to show that a
new source for backscatter modulations will result when the polarization properties of
the scattering matrix are utilized specifically to sense wave tilts occurring in the
azimuthal direction. The improvement in the fidelity of 2-D wave spectra created
using optimal polarizations was investigated using a RAR ocean imaging model. The
new method was derived from observations made on the properties of ocean
polarization signatures. The new imaging technique reported here utilizes linear co-
polar signature intbrmation to maximize a microwave instrument's sensitivity t_)r
azimuthally traveling waves.
Resolved wave tilts create a modulation of the cell polarization orientation
which, in turn, modulates the backscatter intensity in an image. Critical to the success
of the new techniques is the property that ocean linear co-polar signatures have
regions which are highly sensitive to wave tilts. When an image is processed using a
polarization orientation which maximizes this sensitivity, tilt perturbations due to
azimuthally-trave[ing ocean waves have a large effect on the backscatter intensity.
This polarization orientation modulation acts as a new source-term fl_r the
overall microwave modulation transfer function. The magnitude and phase of this
modulation have been quantitatively evaluated. The determination of the magnitude
was demonstrated by making polarization gratings in P-band SAR images of the
ocean. The polarization modulation transfer function contribution has also been
57
evaluated using signatures developed from a theoretical tilted-Bragg model. The
magnitude of the polarization modulation contribution to the MTF is equal to 2-5 formid-range incidence angles. The effect is in phase with the tilt-modulation and has aphase of 90 ° relative to the hydrodynamic term.
2.0 APPLICATION TO WAVE SPECTRAL MEASUREMENTS
A model (Lyzenga, 1988) for the radar imaging of ocean waves has been
modified to include both the amplitude and the phase effects of the new polarizationmodulation source term. Results using this model indicate that undesirable effects such
as distortion and spectral-splitting of the actual spectrum may be greatly reduced forthe case of azimuthally traveling waves.
..,.,E
Z
,q
_-02
e_
z
lIAR MODEL WAVENUMBER SPFX_I'RUM
l [ - L ,
1 i i , .
-02 oo 02
AZIMUTII DIRECTION WAVF2qUMBER - (m "i)
(A)
(B)
'g
i O2
!oo-0 2
RAR MODEl. WAVENUMBER SPECTRUM
i • t • v
' olo ol2
AZIMUTII DIRECTION WAVENUMBER - (m -_)
Figl(A-B): Conventional Model RAR spectrum (A), and the Model RAR spectrummodified to include the new polarization term (B). The spectrum of (B) is similar in
shape to the input waveslope spectrum, whereas the spectrum of (A) exhibits spittingand distortion about the ,I,=0 direction
Figure 1 shows comparisons made between conventional RAR spectra, and
spectra developed with the new polarization modulation term included. Fig. l(A)
shows the conventional RAR waveslope spectra, with the dominant wave (k= lOOm,
significant waveheight h,=3m ) traveling in the azimuthal direction (qb=O°). Severe
58
spectral splitting (abnormal four-lobed pattern) and distortion occurs because of the
roll-off of the transfer function as the direction approaches ,1,=0. Fig. I(B) includes a
polarization modulation term of magnitude 3.0 in phase with the tilt modulation, but
having a cos_ dependence. Fig. I(B) indicates that polarization modulation MTF
contributions are capable of compensating tbr distortions in the RAR spectra which
are due to azimuthal roll-off of the modulation transfer function.
3.0 POLARIMETRIC SIGNATURE MODULATION
A JPL AIRSAR image (20 July 1988) of azimuthally traveling Pacific swell
occurring off the coast of San Francisco was used to apply the new techniques to SAR
images rather than RAR images, or models. The wave-induced intensity modulations
in this image were strong and essentially uni-directional. The modulations were likely
produced by velocity-bunching effects. A first attempt to enhance the wave
modulations by changing the linear polarization through a range of values did not
produce any significant changes. This was probably due to the strength of the
velocity-bunching effects and a rmsmatch in phases between this effect and the
polarization modulation. The original technique was then modified to detect changes in
location of the polarization signature rather than intensity modulations. A strip image
parallel to the azimuthal direction was tbrmed which had pixel averaging (20) in the
range direction and (2) in the azimuthal direction. Polarization signatures were tbrmed
tbr 350 cells formed parallel to the direction of wave propagation. Gates were set at
the peak at other points in the signature to sense wave-induced movement parallel to
the Orientation axis of the signature. Fig 2 shows a power waveslope spectrum of the
strip, this SAR wave spectrum shows a strong peak at a wavenumber k=.032 m 4
indicating the presence of a dominant wave of 192 m wavelength.
*E
1 5x105 l
1.0xlO 5/
5.0x104
P-band SPAN Image Spectrumi I
0.00 0.05 0.10
Wavenumber (m- 1)
10.15
Fig. 2: SAR intensity waveslope spectrum lbr the image strip.
59
spotio,I o_ Modulation4.0xt07 • ,
:q.Ox10 7
I 2.0xtO
OF V! , r I " ,V • l
0.00 0.(36 0.10 0.15
[m-q
Fig. 3: Polarization orieatation modulation spectrum for the
same azimutbally<raveling waves.
Fig. 3 givesa polarizationorientationmodulation spectrum for the stnp
which, agauz,has a strongpeak atthe same wavenumber position.The signatures
used to produce Fig,3were allnormalized tounityto minimize any intensity
modulation.The peak isproduced by signaturelocationchanges due to wave slopes.
This prelimimu'ystudy indicatesthatincaseswhere changes in the polarization
signaturemorphology are slight,slopeinducedorientationchanges can be detected-
and possiblymeasured. The techniqueworks for theocean (orselectedland areas)
where the signatureissharplypeaked and relativelyconstant.
4.0 CONCLUSIONS
A polarim_tnc technique for improving the visibility of waves, whose
propagation direction has an azimuthal component, in RAR or SAR images has beeninvestigated. The technique shows promise as a means of producing more accurate 2-
D polarimetric RAR ocean wave spectra. For SAR applications domination by
velocity-bunching effects may limit its uselhlness to long ocean swell. A modification
of this technique involving measurement of polarization signature modulations in the
image is useful for detecting waves in SAR images and, potentially, estimating RMSwave slopes.
REFERENCES
Alpers, W.R., D.B. Ross, and C.L. Rufenach, 1981, "On the Detectability of OceanSurface Waves by Real and Synthetic Aperture Radar', J. Ge_phys. Res., vol.86,pp.6481-6498.
Schuler, D.L., J.S. Le_, and K. Hoppel, 1993, "Polarimetric SAR Image Signaturesof the Ocean and Gulf Stream Features", accepted for publication in IEEE Trans.Geosci. and Remote Sensing.
Lyzenga, D.R., 1988, "An Analytic Representation of the SAR Image Spectrum tbrOcean Waves', J. Geophys. Res., vol.93, pp. 13859-13865.
60
N95- 23954
AN IMPROVED ALGORITHM FOR RETRIEVAL OF SNOW WETNESS
USING C-BAND AIRSAR
Jiancheng Shi, Jeff Dozier and Helmut Rott
Center for Remote Sensing and Environmental Optics (CSL/CRSEO)
University of California, Santa Barbara, CA 93106, USA
and
Institute for Meteorology and Geophysics
University of lnnsbruck, Innrain 52, A-6020 Innsbruck, Austria
INTRODUCTION
Backscattering measurements by SAR from wet snow covered terrain are affected by
two sets of parameters: (1) sensor parameters which include the frequency, polarization, and
viewing geometry, and (2) snowpack parameters which include snow density, liquid water
content, particle sizes and shapes of ice and water, type of the correlation function and its
parameters of surface roughness.
The identified scattering mechanics of wet snow-covered terrain from the model pre-
dictions and measurements of the polarimetric properties (Shi et al, 1992) shows that the
first-order surface and volume scatterings are dominated scattering source. We can con-
struct, in general, the backscatter model of wet snow-covered terrain with two components:
= (1)
where a is the backscattering coefficient, pp indicates polarization. The subscript t, s,
and v represent the total backscattering, the surface backscattering from the air-snow inter-
face, and the volume backscattering from the snowpack, respectively. The relation between
backscattering and snow wetness is controlled by the scattering mechanism. When the
surface is smooth, volume scattering is the dominant scattering source. As snow wetness in-
creases, both the volume scattering albedo and the transmission coefficients greatly decrease.
This results in a negative correlation between the backscattering signals and snow wetness.
When the surface is not smooth, increasing snow wetness results in greatly increased sur-
face scattering interaction and surface scattering becomes the dominant scattering process.
Therefore, a positive correlation between the baekscattering signals and snow wetness will be
observed. The characteristics of above relationships makes it difficult to derive an empirical
relation from field measurements and a physical based algorithm is needed for a large region
snow wetness estimation.
Our previous works (Shi et al., 1993) indicated that the ratios of a _ to o"nh and
(7whh to a hh could be used for snow wetness retrieval at C-band. The development of
the inversion model was based upon the relations of the first-order surface and volume
backscattering model predictions. The small perturbation model was used to predict therelations between snow wetness and above ratios, which are independent of surface roughness
and only dependent on the local incidence angle and snow dielectric property. Our resent
study (Shi et al,. 1993) showed an over-estimating snow wetness by this method. The tested
results had an average relative error about 23 percent and the maximum error could research
50 percent. This is because the small perturbation model can only be applied to smooth
surface. The accuracy of this model is generally within 2 dB for surfaces with standard
deviation of surface height less than 0.1 of the wavelength and small surface correlation
length (kl less than 3) (Chen et al., 1988).
61
Due to the surface roughness parameters of most of natural surface are outside the range
of the valid conditions for the Small Perturbation and Geometric Optical models, application
of these surface scattering models are greatly limited to certain types of the surface roughness
conditions. The recently developed Integral Equation Model (IEM) (Fung et al., 1991 and
1992) allows a much wider range of the surface roughness conditions. However, it does not
allow to apply this model directly to infer geophysical parameters because of the complicity
of this model and the limited independent observations provided by SAR measurements.
This study shows our continue efforts on developing and testing the algorithm for re-
trieval snow wetness using C-band JPL AIRSAR data. We show (1) a simplified surface
backscattering model particularly derived for wet snow physical conditions from the numer-
ical simulations by IEM model, and (2) snow wetness retrieval model test and comparisonwith ground measurements using C-band AIRSAR data,
INVERSION MODEL DEVELOPMENT
The volume backscattering coefficient is a function of the permittivity and the volume
scattering albedo (depending on snow density, wetness, particle size, size variation and
shape). Under the spherical grain or random oriented particles assumption, the relationship
for the first-order volume backscattering signals of VV and HH polarizations can be alsoobtained:
- TL(0,,c ) (2)
where T_ and T_h are double pass of the power transmission coefficients.
The surface backscattering is a function of the permittivity of wet snow (depending on
snow density and wetness) and the roughness of the air-snow interface which is described by
the auto-correlation function of random surface height, the standard deviation of the surface
height, and the correlation length. Due to complicity of IEM model and the limited number
of independent observations from the polarimetric SAR, we need to minimize or combine
these factors in order to develop an algorithm of measuring snow wetness.
Using IEM model, we simulated surface backscattering coefficients of a,_ and a,_hhat C-band for the possible snow wetness and surface roughness conditions. The simulated
backscattering coefficients cover the ranges for snow wetness from 1 percent to 13 percent,
for the incidence angle from 25 ° to 70", for the standard deviation of random surface height
from 0.1 mm to 15 mm, and for the surface correlation length 0.5 cm to 25 cm. Through
statistical analysis, we found a simplified form for the backscattering coefficients
hA [ 1.12Sn ] '.2_' = I_hhl2 L(o.Ti7-sR)J (3)
: [ 1.06sR ] ,2cr ['(0.17_ + SR. (4)
[otvv]21.05SRcos(8i)
where c%v and ahh are same as that for the small perturbation model and given in (Tsang
et al,. 1985). The Sn is the surface roughness parameter, which is Sn = ks2Wcos2(8). W
is the Fourier transform of the power spectrum of the surface correlation function.
As predicated by the models of the first-order surface and volume backscattering, the
correlation coefficient in H and V channel is perfect correlated. The real part of the cross
product of VV and HH complex scattering elements, Re[S_tvSthh*], can be related to the
surface and volume backscattering coefficients by
= = + (6)
62
Theratioof a_hh/avnhcanbewrittenas
__hh Re[T:_hh.]DTv(O_,¢_)- % -- (7)
Using Equations (1) to (7), we can derive two inversion formula for snow wetness re-
trieval as
DTVO._v _vvhh vv vvah (8)--a t = DTVO" s --as
and
DTa hh - cr7_ -- DTCrh n -- _r__ (9)
In Equation (8) and (9), there are only two unknowns: the snowpack permittivity and
the surface roughness parameter SR. Therefore, we can obtain the estimations of snow
wetness and surface roughness parameter by solve the equation (8) and (9) simultaneously.
COMPARISON WITH AIRSAR MEASUREMENTS
The data used in this study for testing the algorithm are from the NASA/JPL airborne
imaging polarimeter overflying the central part of the Otztal test site on June 25, 1991. The
experiment consisted of three flight passes at a flight altitude about 10,600 meter a.s.I. Two
of the flight lines were aligned in E-W direction, shifted by 4 km. in latitude. Another was
collected from an W-E flight line, resulting in opposite look direction.
At the time of the radar survey the snow cover was wet at all elevation zones. In ground
sampled data, the liquid water content in the top snow layer (5 cm), obtained from average
values at 0 and 5 cm, was ranged from 4.0 to 7.2 percent by volume. Snow grain radii were
from 1.0 to 2 mm in the top snow layer. The snow densities and depths (over glacier ice)
ranged from 460-530 kg m-a and from 40-205 cmrespectively. In addition to snow physical
parameters measurements, surface roughness was measured by a lasermeter. The standard
deviations of surface height were 0.1-0.7 cm; correlation lengths ranged 1-23 cm.
To test the algorithm for measuring snow wetness over a large area, snow-covered area
map was first obtained and non-snow-covered area was masked. Secondly, the stokes matrix
for a given pixel was determined by the mean value within a 3 x 3 window in order to
reduce the effect of image speckle. Figure 1 shows two maps of the inversion-derived snow
wetness, which are derived from two images with E-W flight passes. The image brightness
is proportional to the snow wetness by volume. The black region is non-snow-covered area.
At most of the lower elevation region, the inferred liquid water content of the top snow
layer was in the order of 5 to 7 percent by volume. It decreases to 1 or 4 percent at the
higher elevations. This agrees well ground regional conditions. Both snow wetness maps
derived from different incident angle showed a consistent results within 2 percent. Figure 2.
shows the comparisons between the field measurements and the SAR derived snow wetness
for the locations where the ground measurements were available. The line indicates where
the snow wetness is exactly same from the ground and SAR derived measurements. The
measurements above and below this line indicate an over-estimation and under-estimation,
respectively. The relative error was within 25 percent from all measurements. In overall, the
algorithm performed well and provided a consistent results, less than 2 percent for absolute
values, at different incidence angles. The magnitude of the error is within the range we
expected. Since the algorithm performed at different incident angle produces the consistent
result, it is also possible to calibrate the algorithm.
CONCLUSIONS
This study shows recent results of our efforts to develop and verify an algorithm for
snow wetness retrieval from a polarimetric SAR. Our algorithm is based on the first-order
63
scatteringmodelwithconsiderationof bothsurfaceandvolumescattering.It operatesatC-bandandrequiresonlyroughinformationaboutthe icevolumefractionin snowpack.Comparing ground measurements and inferred from JPL AIRSAR data, the results showed
that the relative error inferred from SAR imagery was within 25 percent. The inferred snow
wetness from different looking geometries (two flight passes) provided consistent results
within 2 percent. Both regional and point measurement comparisons between the ground
and SAR derived snow wetness indicates that the inversion algorithm performs well using
AIRSAR data and should prove useful for routine and large-area snow wetness (in top layer
of a snowpack) measurements.
Chen,
Fung,
Fung,
Shi,
Shi,
Tsang,
REFERENCES
M. F. and A. K. Fung, 1988, "A numerical study of the regions of validity of the
Kirchhoff and small perturbation rough surface scattering models," Radio Science, vol.
23, pp 163-170.
A. K. and K. S. Chen, 1991, "Dependence of the surface backscattering coefficients on
roughness, frequency and polarization states", Int. J. Remote Sensing, vol. 13, no. 9,
pp. 1663-1680.
A. K., Z. Li, and K. S. Chen, 1992, "Backscattering from a randomly rough dielectric
surface," IEEE Transactions on Geoseience and Remote Sensing, vol. 30, no. 2, pp.356-369.
J. and J. Dozier, 1992, "Radar response to snow wetness," Proceedings IGARSS '92,vol. 2, pp. 927 929.
J., J. Dozier, and II. Rott, 1993, "Deriving snow liquid water content using C-band
polarimetric SAR", ill press Proceedings IGARSS '93.
L., J. A. Kong, and R. T. Shin, 1985, in "Theory of microwave remote sensing," Wiley
publication, New York, pp. 108. 1992.
Figure 1. C-band SAR derived snow wetness maps.
c
_ p, ..
Clg
E
g_D
i
4 6 8 10
Ground measutemenls
55 60 65
Incidence angle
Figure 2. Comparison of ground measurements with SAR derived snow wetness.
64
N95- 23955
POLARIMETRIC RADAR DATA DECOMPOSITION AND
INTERPRETATION
Guoqing Sun t and K. Jon Ranson 2
I Science Systems & Application, Inc. 5900 Princess Garden Parkway,Suite 300, Lanham, MD, 20706, USA.
2 Biosphcric Sciences Branch, GSFC Greenbelt, MD 20771, USA.
I. INTRODUCTION
Significant efforts have been made to decompose polarimetric radar data into
several simple scattering components. The components which are selected because of
their physical significance can be used to classify SAR image data. If particular com-
ponents can be related to forest parameters, inversion procedures may be developed to
estimate these parameters from the scattering components.
Several methods (van Zyl, 1989; Freeman and Durden, 1992; van Zyl, 1992)
have been used to decompose an averaged Stoke's matrix or covariance matrix into
three components representing odd (surface), even (double-bounce) and diffuse
(volume) scatterings. With these decomposition techniques, phenomena, such as
canopy-ground interactions, randomness of orientation and size of scatterers, can beexamined from SAR data.
In this study we applied the method recently reported by van Zyl (1992) to
decompose averaged backscattering covariance matrices extracted from JPL SAR
images over forest stands in Maine, USA. These stands are mostly mixed stands ofconiferous and deciduous trees. Biomass data have been derived from field measure-
ments of DBH and tree density using allometric equations. The interpretation of the
decompositions and relationships with measured stand biomass are presented in this
paper.
2. DECOMPOSITION
van Zyl (1992) showed that for azimuthally symmetrical terrain in the monos-tatic case, the average covariance matrix of backscattering can be decomposed as:
[TI = C _q = _iki ki + (1)
0 i=1
where C:'<ShhShh*>, p:<ShhShv*>, _:2<ShvShv *>If , _:<SvvSvv *>If. The
_i,i=1,2,3 are the eigenvalues of [T]. k,,i=1,2,3 are the corresponding eigenvectorsand + means adjoint. Since the eigenvectors are unitary vectors and the sum of the
eigenvalues equals the total power of the backscattering, _.x,_.2,_.3, are the backscatter-
ing powers contributed by odd, even and diffuse backscattering components, respec-tively. We also note that the _.3 is exactly the backscattering power at cross-
polarizations, i.e. 2<ShhSh_,*>. In terms of backscattered power, this algorithm
decomposes the power from co-polarized returns into odd and even scattering com-
ponents. For those targets with p=0, depending on the _>1 or _<1, one of the two
eigen values (either _, or _-z) equzds the HH return and the other the VV return.
When _= 1, the odd and even scattering components are equal.
65
3. RESULTS AND DISCUSSION
3.1. Decomposition and Forest Biomass
Figure 1 presents scatter plots of total above-ground fresh biomass of 47 forest
stands versus 3,1,_,_. 3 and _r° at HH, HV and VV polarizations at L band. Table 1
summarizes the correlation coefficients for scattering components and measured
biomass. Comparing the two plots in the third row demonstrates that _.3 (diffusescattering) is identical to the sum of the cross-polarization backscatter cross sections.
The X2 (even scattering component) has higher correlation with biomass than the odd
scattering component. In the first-order backscatter models, the odd scattering is from
crown backscattering and direct backscattering from ground surface. If the canopy is
dense and tall, crown backscattering will be the major source and the odd scatteringshould have higher correlation with forest biomass. This is not obvious from the data
shown in Figure 1 or listed in Table 1.
3.2. Decomposition and Forest Classes
Table 2 lists the decomposition results of several classes at C, L, and P bands.
Generally, the odd scattering is always the major component. The even scatteringcomponent is higher for forest stands with large and dense trees at L and P bands.
The higher entropy values of dense forest stands at L and P bands show a high degreeof disorder (randomness) of scatterers. At C band, except for Bog and Red Pine sites,all sites have the similar entropies.
3.3. Decomposition of modeled Scattering
Backscatter models (Sun, 1990) were used to simulate backscattering from the
47 stands. The tree density and size for a stand were from field measurements, but
trees were assumed to be pure hemlocks and the ground surface to be a rough surfacesimilar to an old cut area near these stands. The decomposition of SAR data and
modeled scattering matrices at L band were compared in Figure 2. The simulated
components have good correlation with biomass. Though the comparison between
SAR and simulated data is crude, it seems that model gives reasonable results in termsof even and diffuse scattering but not for odd scattering.
4. SUMMARY
The decomposition method partitions the co-polarization return into odd and
even scattering components. The partition depends on two parameters, i.e. p and
only. It helps to classify radar polarimetry return into general groups of scatteringbehavior.
The HV backscatter or diffuse components has the best correlation with forestbiomass.
Comparing to HV backscatter, HH and VV backscatters have higher signal to
noise ratio and are desirable for developing an inversion algorithm for forest parameter
estimation. More works, however, need to be done to separate scattering componentsheavily influenced by ground surface from the co-polarization signatures.
REFERENCES
Freeman, A. and S. Durden, 1992, Fitting a three-component scattering model topolarimetric SAR data, Summaries of the Tlu'rd Annual JPL Airborne Geosci-
ence Workshop, Volume 3, AIRSAR Workshop, Ed. J. van Zyl, pp. 56-58.
Sun, G., 1990, Radar Backscatter Modeling of Coniferous Forest Canopies, Ph. D.
Dissertation, University of California, Santa Barbara. 126 p.
66
vanZyl, J. J., 1989,Unsupervisedclassificationof scatteringbehaviorusingradarpolarimetrydata,IEEE Transactions on Geoscience and Remote Sensing, Vol.
27, No. 1, pp. 36-45.
van Zyl, J. J., 1992, Application of Cloude's target decomposition theorem to polar-
imetric imaging radar data, Proceedings of SPIE, Vol. 1748, Radar Polarimetry,Eds. !1. Mort and W. M. Boerner, 23-24 July 1992, San Diego, California, pp.184- 191.
d3
"t3c.- o
.9,,J
Od
E_
'gZ-9,_J
"O t'_
eO ,v--
-Jm,-
]. "..%7 "ee
4"7
• .....
• J .0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5
Log1 0(Biota ass) Log10(Biomass)
o•
0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5Log10(Biom ass) Log10(Biomass)
"1"
-T-CO
"O 't-
• *. . "s.l'
• g* s
N
• - o•133
t'Xl,,_. I •
¢-- ,
..O .,-.- '1
-...,._."
0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5Log10(Biomass) Log10(Biomass)
Figure 1. Comparisons of SAR original HH, VV and HV backscattering with decomposed scattering in
terms of their relation to total above-ground fresh biomass.
Table 1. List of correlation parameters to biomassfor variables in Figure 1 (y = bo+blx).
b0 bl R z F-value R.S.E_,] 2.470 0.193 0.281 17.54 0.436X,z 3.458 0.237 0.601 67.79 0.3245_,3 3.260 0.201 0,680 95.73 0.291VV 2.971 0.202 0.316 20.80 0.425HH 2.843 0.225 0.455 37.59 0.379
67
.--_O
...a"T
_j ,
E_
.-JCO
__ CO
Table 2. Decomposition of SAR data at C, L and P bands.
Site X1(%) _2(%) _.3(%) Total Power En_opyC Band
Grass 57.94 23.09 18.97 0.2176 0.8829
Bog 78.15 12.02 9.83 0.5508 0.6147
Regen 58.71 23.16 18.13 0.3183 0.8748
Clear 61.84 20.34 17.82 0.4676 0.8451
Aspen 57.73 20.49 21.78 0.4632 0.8865
Mixed 63.55 19.74 16.71 0.4382 0.8259Hemlock 62.64 20.73 16.63 0.4385 0.8352
Red Pine 49.38 21.59 29.03 0.2596 0.9452
Spruce 60.71 20.01 19.28 0.5368 0.8576:L Band
Grass 81.88 10.56 7.56 0.0700 0.5429
Bog 84.00 8.03 7.97 0.3086 0.5011
Regen 53.14 25.56 21.30 0.1900 0.9230
Clear 61.32 21.48 17.20 0.3105 0.8492
Aspen 42.54 30.73 26.73 0.3921 0.9820
Mixed 46.42 29.89 23.69 0.3746 0.9634
Hemlock 45.69 28.27 26.04 0.4263 0.9698
Red Pine 44.06 27.68 28.24 0.5585 0.9774
Spruce 50.73 27.13 22.14 0.5488 0.9394
P Band
Grass 89.64 6.87 3.49 0.0848 0.3631Bog 88.48 6.32 5.20 0.1622 0.3975Regen 60.42 25.01 14.57 0.1516 0.8480
Clear 63.24 22.16 14.60 0.2245 0.8230
Aspen 52.37 23.18 24.45 0.3627 0.9303Mixed 49.13 30.07 20.80 0.3474 0.9440
Hemlock 48.83 28.37 22.80 0.4596 0.9508Red Pine 58.95 24.64 16.41 0.6692 0.8677
Spruce 53.43 25.28 21.29 0.4608 0.9210
• o •
.-...:.,._e ,_,t I,e•
•o
00 05 10 15Log10(Biomass)
ee •e
00 05 10 15Log10(Biomass)
u
,o _o ••
8
E,'- °
mo
v,
tU'7,
_'7, •..d
• J
• ° • •-'1"11, • .% • •
• . • • • °
• _o e e • •
-10 -8 -6 -4SAR L-band Odd (dB)
"'" ".s"2::t
-14 -12 -10 -8SAR L-band Even (dB)
A
. :_- t"xl I " •
"O',¢ II... 1
J;_(.o | • _•
_cOI •
U) -1800 05 10 15 -16 -14 -12 -10 -8Log10(Biomass) SAR L-band Diffuse (dB)
Figure 2. Decompositions of modeled backscattering and comparison with SAR data.
68
N95- 23956
SYNERGY BETWEEN OPTICAl. AND MICROWAVI: I{EMOTESENSING TO DERIVF SOIL AND VI,_GI,:TATION PARAMETERS
FROM MAC EUROPE 91 EXPERIMF.NT
TACONET 1, O., BENALLEGUE 2, M., VIDAL 3, A., VIDAL-MADJAR 1, D.,L. PREVOT4,M. NORMAND 2
1 Centre de Recherches en Physique de l'Envimnnement Tcn'esu'e et Plan_taire10-12, avenue de l'Europc 78140 VELIZY (Frzmce)
2 CEMAGREF, Division 11ydrologie- BP 121 F-92185 Antony Cedex (France)3 CEMAGREF/LCT, 361, rue J-F Breton .B.P. 5095 34033 Montpellier Cedex 1
4 INRA, Bioclimatologie, BP 91, 84140 MONT[:AVET (France)
I. INTRODUCTION
The ability of remote sensing for monitoring vegetation density and soil moisturefor agricultural applications is extensively studied. In optical bands, vegetation indices(NDVI, WDVI) in visible and near infrared reflectances are related to biophysicalquantities as the leaf area index, the biomass. In active microwave bands, the quantitativeassessment of crop parmneters and soil moisture over agricultural are_ks by radarmulticonfiguration algoritms remains prospective. Futhermore the main results are mostlyvalidated on small test sites (Ulaby et at. 1984), but have still to be demonstrated in an
operational way at a regional scale.
In this study, a large data set of radar backscaltering has been achieved at aregional scale on a french pilot watershed, the Orgeval, ahmg two growing seasons in1988 and 1989 (mainly wheat and corn). The radar backscattering was provided by theairborne scatterometer ERASME, designed at CRPE, (C and X bands and 1{11and VV
polarizations). Empirical relationships to estimate water crop ,and soil moisture over wheatin CHH band under actual field conditions and at a walershed scale are investigated.
Therefore the algorithms developped in CHH band ,are applied for mapping the surfaceconditions over wheat fields using the AIRSAR and TMS images collected during theMAC EUROPE'91 experiinent. The synergy between optic:d and microwave bands is
analysed.
2. THE ORGEVAL CAMPAIGNS ' 88,89 and 91
The characteristics of the scatterometer ERASMF. is p,escnted in Table 1. The
French test site is the Orgeval experimental watershed of 5 by 5 km 2 (France), mainly
covered by wheat and corn with silt loamy soils. 49 fields of wheat and 12 fields of cornare identified. During 1988 (AGRISCATF), the scatteromctcr L:RASME was performedalong a 17 km axis through the basin during June and July with 17 test fields (wheat andcorn). In 1989, it was performed along 14 crossed-axis cove,ing the watershed (21 testfields) for every month from March to December 1989.
During the Mac-Europe Campaign'91, flights of the NASA airborne sensors (themultispectral radiometer qTvlS, the synthetic aperture radar AIRSAR) have been done,simultaneously with flights of ERASME.The two intensive periods were two weeks, thefirst mid June and the second mid July. 2 test fields (wheat and corn) were instrumented.
Only two enough clear TMS images ,are available, the 17 July and the 22 July. Theconcomittant AIRSAR image is the 16 July.Ground u'uth measu,cmcnts related to soiland vegetation are soil moisture, soil profiles with a 2 meter pin-profiler, leaf area index,
crop height, biomass and water content.
3. EMPIRICAL RELATIONS IN CHH BAND
Considering the distribution of radar cross sections at 20 and 40 degrees ofincidence angles with the vegetation water content (Figures 1 and 2), it appears that thebehaviour of radar backscattering can be divided in two cases, low vegetation cover andhigh vegetation cover. For vegetation water conteut lower than 2, the cross sections forboth 20 and 40 degrees are highly variable. The vegetation water content is not the
69 ORIGINAL PAGE IS
OF POOR QUALFrY
Table 1 Radar Characteristics of ERASME
Type
Frequencies
Transmitted
Power
Modulation
Antenna Axis
Position
RangeResolution
Antenna Range
Altimeter
Antenna
Pixel Size
Accuracy
Forward looking FM/CW
5.35 GHz and 9.65 GHz
11.2 dBm at C band
20 dBm at X band
Sawtooth, 3ms of period
23 ° , 38 ° and 45 °
0.97m at 23 °
1.30m at 38 ° and 45 °
+10 ° in elevation
+2 ° in azimuth
Nadir Hom
20m by lOm
ldB +7 ° apart the axis
Figure I- Radar backscattering at 20 degree of incidencefrom Orgeval'88-89 related to vegetation water contentin kg/m 2 and classilied with soil moisture in cm3/cm 3
ORGEVAL 88-8g
Mv obs
kg/m2
5
4
3
2
1
0-14
•", mvinf17% wg<17%
• mvsup29% wg>29%
i !X rn lhvint 17%<wg<2g%
! i J i
_& iA
-12 -10 -8 -6 -4 -2 0,,Ig 20* dB
dominant parameter. The relevant parameter for radar cross sections is the soil moisture.
For vegetation water content Mv higher than 2, it appe,'us a linear relation between the
radar cross sections and the vegetation water content. The radar backscattering is decrasingwith increasing crop coverture. The soil moisture is no more an influent parameter.
3.1 Case of low vegetation cover
The relation of cross section with soil moisture Wg for bare soil is establishedusing data points with the lowest values of vegetation cover (Mv<lkg/m2). A linearrelation is obtained at 20 and 40 degree as proposed by Ulaby (1978):The 2 obtained relations at 20 and 40 degrees are:
c0soil(20)= -12.1+0.18 Wg (1)
c0soil(40)= - 13.8+0.14 Wg (2)
with comparable slopes and a shift of about 2dB.This simple linear relation is obtained only for wheat culture. Linear relations betweenradar cross sections and soil moistures ,are no longer available for bare soils in the cases ofcorn sowing or of ploughs.
Linear relation between radar cross sections and soil moistures _Lreobtained with
low vegetation cover, (Mv< 3 kg/m2) ( Figure 3). The relations at 20 and 40 degrees ,are:
c0soil(20)= - 15.6+0.29 Wg (3)
c0soil(40)= - 14.9+0.14 Wg (4)
The cloud model has been adjusted over wheat, taking only the attenualed part by thevegetation:
c0= t2 c0soil and t2 = exp(-2B My/cos0) and cr0soil in dB= C1 - C2 0 + D Wgwith 0 the incidence angle.
t_0 in dB= -8.69 B My/cos0 -C1 - C2 0 + D Wg (5)
The adjusted coefficients are:B=0.09,Cl=-8.32.C2 =0. 147,1)=0. 193 (comp_u'able withresults of Pr6vot et al. 1993).
3.2 Case of high vegetation cover
For dense canopy (Mv>3kg/m2), the soil moisture is no mo,'e a relevant
70
Figure2-Radarbackscatteriugat40degreeofincidencefromOrgeval'88-89relatedtovegetationwatercontentinkg/m2andclassifiedwithsoilmoistureincm3/cm3
Zx mvinf17% hv<17%ORGEVAL 88-88
• mvsup29% hv>29%
X mvhv nt 17%<hv<29%Mv obs 5 _r _ t t r i i
kg/m2 _ A
3
AAAAI k AA
1 _r, A _ A,,',"ax , ,j
0 t t I L I-11 -16 -14 -12 -10 -8 -6
sig 40 ° dB
I:igurc 3: [.me;u regression between radar cross sectionand soil moisture over wheat for low cover canopy
(vcgctation water content<3 kg/m 3)
-4
sicJ 20 ° 0
dB
-2
-4
-6
-8
-10
-12
ORGEVALS8-89
sig20,,-15.5$.0.290wg R=0.777
-i-:
i ÷i :........... _ ........... i........... _" + Ii....... 5;- ........... -
i_.+ i :..... _.............. , .............. ; .......... _.'_......:t.-.... _.............:: 2..÷÷ i :
+ + ! i ! -
I I I
15 20 25 30 35 40
wg % cm3/cm3
parameter to parameterize the ra&u" cross section (see Figures 1 and 2). The cross sectionsat 40 degrees are related quasi linearly to vegetation water content. The radar backscatteringis attenuated when the foliage density is increasing (Prtvot et al., 1993). At 20 degrees,radar cross section is highly variable for the same foliage density indicating that otherstructural parmeters of canopy have to be accounted. Experimental linear relation betweenradar cross section and vegetation water content can be prolx_sed from the data set.
At 40 degrees, Mv=-0.35o0dB-1.44 (6)
But as the dependance of the cross sections with the soil moistme disappears, theformulation given by the attenuated part of the cloud model fitted for Mv<3kg/m 2 is nomore available.
4. APPLICATION TO THE ORGEVAL'91/MAC-EUROPEEXPERIMENT
A map of the 49 fields of wheat over the basin is given in Figure 4. The synergybetween the TMS (17 July) and the AIRSAR (16 July) images is investigated. The TMS
image has been radiometrically calibrated and corrected from atmospherical diffusion invisible/near infrared bands. Approximate calibration of LAI (le_d"area index) versus NDVI(Normalized Vegetation index) and vegetation water content _'Iv) versus LAI are obtainedover- the reference field:LAI=- 4.1+11.4 NDVI (7)Mv=2.32+0.25 LAI (8)
Spatial variations of NDVI and Mv over the wheat parcels _u'e derived (Figure 5).
As the magnitude of the estimated Mv are between 2.8 and 3.4 kg/m 2, only a map of thevegetation water content can derived froln the radar cross section of the AIRSAR images (2 images around 45 degree of incidence with 2 flight directions, perpendicular and parallelto the average rows direction in the watershed). The algoritlun related radar cross sectionand Mv (Eq. 7) has been used. An intercalibration between the AIRSAR and ERASMEdata is done before, as the AIRSAR data are systematically lower of about -3.5dB.Therefore AIRSAR data are decreased of -3.5dB and alter of +ldB to approximate the cross
sections at 40 degrees needed in Eq. 7. The value of-1 dB has been calculated fromstatistical relation between the ERASME data at 40 and 45 degree.
It has been noted that the derived vegetation (LAI,Mv) pa,amcters frommicrowave algoritms are lower and more scattered than those deduced from the optical
vegetation index. Nevertheless the accuracy of relationships using optical vegetationindex is also to be assessed.
71
Figure 4- M_ap_of__wheatfields over the Orgeval
5- CONCLUSION
Figure 5: Spatial variations of NDVI and deduced
vegetation water content My (kg/m 2) over wheatfields. "
ORGEVAL91
--.re.--NOV[ TUS _r JULY
0.y ............. !............... _............... _................ .y........... __o,I.............i...............i..............i..............i.............
0.7_
0.6
0.5
: i if I f
0"0 10 20 30
field
_Mv(ndvi)
4
3.6
3.2
2,8
2.4
f40 5'i_
A complementary use of the optical and microwave bands is proposed. Overwheat, the knowledge of the vegetation index values appear necessary to discriminatedense or low vegetadon cover over the wheat fields and choose the adequate microwavealgoritm to derive either the soil moisture, either the water content of the vegetation. InCHH band, radar data at 20 and 40 degree can be used to derive soil moisture for low
cover. Radar data at 40 degree are used to derive water content and are not saturating assoon as optical vegetation index NDVI.
Figure 6: Intercomparison of radar cross section fromAISAR and ERASME
COMPARISON AIRSAR ET ERASME16 JULY 91 ORGEVAL
m
o'O
E_
-5
-7
-9
-1!
-13
-15
sig erasm.:-S.9÷0.$6 slg airsar R:0.7
i...............i
i
S ......... ..............i................-13 -i I -9 -7 -5
a|rsar4S si 9 d8
Figure 7: Simultaneous estimation of Leaf Area Index
and vegetation water content from optical (TMS)andmicrowave (AIRSAR) images.
_Mv(ndvi) +LAIndvi
--6--M V(sig45) 1 "-1_" LAl(sig45) 1ORGEVAL91
SYNERGY FROM TMS AND AIRSAR
35 _ 17(AIRSAR) and 16 July(TMS) --4 5
---_ ! i 43.1 ¢'_, ..... "........ : .......-.................:3.5
2.7 _!' : __:' 2.52.3 i
i !;i:!i!:ii i \/!1.9 ....... !............................ _ ............. _k.- .5
" 1
1.5 I I 0.50 10 20 30 40 50
fieldREFERENCES
Pr6vot, L., M. Dechambre, O. Taconet, D. Vidal-Madjar, M. Normand and S. Galle, 1993,"Estimating the characteristics of vegetation canopies with airborne radar measurements, Int. J.ofRemot. Sens., to be published.
Ulaby,F., T. et al., 1978,"Microwave backscatter dependence on surface roughness, soilmoisture and soil texture: Part l-bare soil, IEEE Trans. Geosci. Electron., vol. 16, no.4,pp.286-295.
Ulaby, F. T., C. T. Allen, G. Eger and E. Kanemasu, 1984, "Relating the microwave
backscatteringf coefficients to leaf area index",Remote Sens. Environ., vol. 14, pp. 113-133.
72
N95- 23957
SAR TERRAIN CLASSIFIER AND MAPPER OF BIOPHYSICALATTRIBUTES
Fawwaz T. Ulaby, M. Craig Dobson, Leland Pierce and Karnal Sarabandi
Radiation Laboratory
Department of Electrical Engineering and Computer ScienceThe University of Michigan
Ann Arbor, Michigan 48109-2122
1. INTRODUCTION
In preparation for the launch of SIR-C/X-SAR and design studies for futureorbital SAR, a program has made considerable progress in the development of an SARterrain classifier and algorithms for quantification of biophysical attributes. The goal of
this program is to produce a generalized software package for terrain classification andestimation of biophysical attributes and to make this package available to the largerscientific community. The basic elements of the SAR terrain classifier are outlined inFigure 1. An SAR image is calibrated with respect to known system and processor gainsand external targets (if available). A Level 1 classifier operates on the data to differentiate:urban features, surfaces and tall and short vegetation. Level 2 classifiers further subdividethese classes on the basis of structure. Finally, biophysical and geophysical inversions
are applied to each class to estimate attributes of interest.
The process used to develop the classifiers and inversions is shown in Figure 2.Radar scattering models developed from theory and from empirical data obtained by truck-mounted polarimeters and the JPL AirSAR are validated. The validated models are used insensitivity studies to understand the roles of various scattering sources (i.e., surface,trunk, branches, etc.) in determining net backscatter. Model simulations of cr° as
functions of the wave parameters ( ;L, polarization and angle of incidence) and the
geophysical and biophysical attributes are used to develop robust classifiers. Theclassifiers are validated using available AirSAR data sets. Specific estimators are
developed for each class on the basis of the scattering models and empirical data sets. Thecandidate algorithms are tested with the AirSAR data sets. The attributes of interestinclude: total above ground biomass, woody biomass, soil moisture and soil roughness.
il ......
• r _tl,,e,
Figure 1. SAR terrain classifier
1
No¢ - Allgorilhm T_llng --
Figure 2. Development and validation process
ORIC !e41 L. ISOE,POOR QUALIFY
73
2. LEVEL 1 CLASSIFIER
After calibration, an SAR image is classified in two stages. The Level 1classifier operates at a pixel level to distinguish urban, tall vegetation, short vegetationand bare surfaces. The classifier has been designed for use with S1R-C/X-SAR. Hence,the data inputs are polarimetric L- and C-band data. JPL AirSAR 4-look data forPellston, Michigan are used to train the classifier. The classifier uses: cr° (hh, vv andhv) at L- and C-bands, the peak co-polarized relative phase difference and texture. TheFORTRAN classifier uses a knowledge-based, binary decision tree to differentiate thefour terrain classes as shown in Figure 3. Figure 4 shows the classification of thetraining image. The classification accuracy is found to be in excess of 90% (see Table 1)for both the training data and independent test areas within the scene. Classifier errorsgenerally involve assignment of short vegetation to either bare surface or tall vegetationclasses. Application of the classifier to other images obtained at a SIR-C/X-SARsupersite near Raco, Michigan yield equally impressive results.
AIR SAR IMAGE l
I'ixel = Urban?iTexture,Ct] Urban
_oPixel = Tall Veg?
[O.ohh(L),Oon,(L)] J Tall Vegetation
[_o
Pixel = Short Veg? _ Short Vegetation[o°,, (C),T...... (L)}
Pixel = Bare Surface ?Ires _ Water
[,_°_(c),_r°h.(L)I _ Bare Snrface_. C,,rou.d
_Short Vegetation
Figure 3. Level 1 classifier design
Table 1. Level 1 Classifigr Res_tlts
Training Areas
True Class
Bare
Classified As Urban Tall Veg Short Veg _Surface
Urban 99.3 0.22 0 (I.06
Fall Veg 0 98.32 0 0
Sl'nn'l Veg 0.50 1.46 94.74 0.87
Barc Surfate 0.20 0 5.26 99.07
Independent Test Areas
Classified As
Urban
"Fall Veg
Short Veg
Bare Surface
True Class
Bare
Urban Tall Veg Short Veg Surface
99.1 0 0.85 0.01
0.1 98.04 2.84 0.01
0.63 1.96 90.77 0.18
0.17 0 5.54 99.79
Figure 4. Classified AirSAR images from Pellston,MI. Urban = white, Tall Vegetation = light grey,Short Vegetation = dark gray, Bare surfaces = black.
-5
-10
-15
-20
-25
-30
.1
,:'_,d_._.... vv
,__=o HV
* " P-band
; " ...... 1_0 ........ 100' ........ 1000
Blomass (tons/ha)
Figure 5. P-band response to forest biomass. Datafrom loblolly pines (Duke Forest) and maritime pines(Landcs Forest).
74
3. LEVEL 2 CLASSIFIER
The Level 2 classifier is used to select appropriate estimator algorithms for a
given pixel. It operates on the premise that multifrequency SAR is sensitive to surfaceand canopy structure (i.e., geometric attributes). AirSAR studies of pines at the LandesForest in Bordeaux and the Duke Forest in North Carolina show a power-law dependence
of 0.0 on aboveground biomass, cr° is found to saturate at biomass levels that scalewith wavelength: P-band = 100 tons/ha, L-band = 50 tons/ha and C-band = 10tons/ha. The P-band response shown in Figure 5 indicates that biomass estimators arefeasible below the saturation level. For the pine forests (excurrent tree form), AirSAR
data indicates a dependence on crown structure only at C-band. Natural forests inparticular are not mono-specie. Examination of AirSAR data from test sites in northernMichigan at Raco and Pellston demonstrates that the different branch and trunk structureof decurrent tree forms yield distinctive biomass relationships at long wavelengths.Hence, it is necessary to subclassify tall vegetation into distinctive SAR structural
categories prior to application of estimator algorithms.
The level 2 classifier for tall vegetation is based upon model results usingMIMICS, a 1st order vector radiative transfer model for closed canopies. Simulations for
grass, shrubs, and excurrent and decurrent tree torms use growth models for canopiesranging from 0 to 200 tons/ha. The model yields modified Mueller matrices from which0.0 and relative phase properties can be calculated for net backscatter as well as the
component source terms. An example of the simulated data is shown in Figure 6.
Model results at P-, L-, and C-bands and 20°< 0 < 60 ° provide a number of
potential classifiers for separation of tall vegetation into shrubs, excurrent trees anddecurrent tree forms. Testing of these classifiers with mixed-specie AirSAR data from
o o
Raco and Pellston, Michigan shows the spectral gradient ( cr;t1 - o)t,2 ) tO be a sensitive
discriminant of tree form. The gradient from P-band to C-band yields the best results asshown in Figure 7. L-band is not a suitable substitute for P-band because of attenuationby the crown layer of branches and leaves.
9.
tXl
L-band, HH
-5 ExcurrentTrees(
-I0 DecurronlTrees _
-15 Green Shrube
-20- Woody Shrubs
-25.
-30_ :D" ...... D ...... Grass
-35.
-40o _ lb ;s 20
Total Biomass (kg/m 2)
Figure 6. MIMICS simulations of backscatter
from various vegetation classes at 0--40 °.
-9,r,.)
0 _
-5
-10
-15'
• , , . I • . • • I • • , , I • • . •
Total Power
Docurrent
I
.... I .... I .... I ....
-10 -5 0 5
P-band (dB)
Figure 7. The spectral gradient is greater forexcurrent tree forms.
75
4. LEVEL 2 MAPPER OF SURFACE PROPERTIES
A suite of estimator algorithms are being designed to operate on each sub-class.
For vegetated regions, the algorithms yield estimates of biomass, stem density, canopyheight, etc. For bare surfaces, estimator algorithms quantify surface roughness and near-surface soil moisture. The surface algorithms have been developed on the basis of truck-
mounted polarimeter data. Semi-empirical formulations using polarization ratios yieldestimates of surface roughness and soil permittivity. Soil moisture is inferred from
permittivity. Some results of this algorithm are shown in Figure 8 for scatterometer data(Oh, 1992). Polarimetric SAR data is required and long wavelengths are preferred (i.e., P-and/or L-band).
5. CONCLUSIONS
An SAR-based terrain classifier and biophysical attribute mapper has beendesigned on the basis of theoretical models and empirical evidence. The level 1 and 2
classifiers take advantage of multifrequency, polarimetric SAR as a structure mapper.Tests of the classifiers on AirSAR data from two distinct sites in northern Michigan atdifferent times of year produce excellent results.
Estimator algorithms for surface attributes have been developed and tested usingscatterometer data. Evaluations of the algorithms using AirSAR data from Chickasha,
OK and Davis, CA are currently underway. Estimator algorithms for canopy biophysicalattributes are in development and will be tested on AirSAR data prior to the SIR-C/X-SAR launch in 1994. The classifier and estimators require polarimetric data at L- and C-bands. The addition of P-band is found to yield superior results for forested areas due toincreased penetration and sensitivity to trunk attributes.
REFERENCES
Oh, Y., K. Sarabandi, and F.T. Ulaby, 1992, "An Empirical Model and an InversionTechnique for Radar Scattering from Bare Soil Surfaces," IEEE Trans. Geo. Rem. Sens.,vol. 30, no. 2, pp.370-381.
5
E
o 0.oInput: (7°hh.Cr _, _, Output: m ,s
t]4,to
o 19oo m¢ _
2o
l O c., o
oo[_ =<_
O0 I I, 20
',.iea> ured
ca,,
rms he_ghl
o
0 o
O2
OI
00'_0 _0 O0
co_l coc_f =09"_ Q
r990m=J_ Oo
Jgqlm¢,i ,""
,'_o,'[3
0 0,,'
, k , _ j
0 1 02 0.3
Measured rn
soil mossmre
04
Figure 8. Results of surface estimator algorith,n.
76ORI_IN,Id.. PAGE IS
OF POOR
N95- 23958
Classification and Soil Moisture Determination of Agricultural
Fields
A.C. vall dell Brock and J.S. Groot
Physics and Electronics Laboratory TNO
P.O. Box 96864, 2509 JG The tlague, Tile Netherlands
1 Introduction
During the Mac-Eurot)e campaign of 1991 several SAR experiments were carried out in the Flevoland
test area in the Netherlands. The testsite consists of a forested and a agricultural area with more than
15 different crop types. The exl)eriments took place in June and July (mid to late growing season).
The area was monitored by tile spaceborne C-band VV polarised ERS-1, the Dutch airborne PHARS
with similar frequency an(l polarisation and the three-frequency (P-, L- and C-band) polarimetric AIR-
SAR system of NASA/JPL. The last system passed over on June 15, July 3, 12 and 28. The last two
dates coincided with the overpasses of the PttARS and the ERS-1. Comparison of the results showed
that backscattering coefficients from the three systems agree quite well(van den Broek and Groot, 1993).
In this paper we present the results of a. study of crop type classification (section 2) and soil moisture
deterininalion in the agricultural area (sect ion 3 ). For these studies we used field averaged Stokes matrices
extracted from the A IRSAR data (processor version :1.55 or 3.56).
2 Classification of agricultural fields
Field averaged Stokes ma.trices contain five non-zero cross products (O'hh:<,._hhS_h>, Gvv=<SvvS_v>,
Ghv_-- <,ShvShv"* >, p= <,5"hhSvv> ), where the last. cross product is complex. The <S_oS_,o,s> products
are zero due to azimuthal symmetry. We use here two classification methods: a Gaussian maximum
likelihood (GML)method which uses the t)olarimetric information directly and the so-called polarimetric
contrast classification (P(:(') method which uses this information lnore indirectly. For the study of crop
type classification we have selected 330 agricultural tields with 8 crot)-type classes (see Table)
crop type #fields crop type #fields
rat)eseed 13 sugar beet 63
grass 41 corn 15
t)otato 86 barley 19
wheat 84 beans 9
2.1 Gaussian lnaximuln likelihood classification
This method deals with feature vectors of arbitrary dilnensionality. We can use single features as C-band
0 y_t,,0 =o oo multi-temporal vectors as X-(o'15/6,o'3/7,cr_ (ERS-I), full polarimetric vectors a.s "'--t " hh ,",_, h,, P) or _ o o
0 0az2/r,_r2s/7) fox' one particular t)ola.l'isation combination. We obtain ensemble statistics fox" every crop-type
class by calculating the mean vector/q=E[X d and the covariance matrix Cij=E[(xi-#i)(xj-#j) t] with E
the expectation value. Next we calculate for every field the distance function D defined by
tloglCl_
to all crop-type classes. The field is assigned to that crop-type class for which this distance is a minimum.
77
100
80 -
60"
¢J40 -
_ .20 -
0
• I .... I I " '
I
I ' I i
153._sl_sy!_2BI]s,312?B., II , , 11, . .
polarimetric
lOOt' I ........ I .... i'
80 .2"60 -
II
,o7 c L ,, p2O - I
• 6._ Re(p)
raa e,. hate) J e;** =,, Irate).., ....,, --",,, p .,...,,)
0 : ?" _ :." I .... I ....
mu]titemporM
.J1
--i
1J
Figure 1: Classification results with the GML (full line) and PCC (dotted line) method using polarimetric
(Fig. la, left) and multitemporal features (Fig. lb, right).
2.2 Polarimetric contrast classification
This method was originally introduced by Kong (1990). It. uses the so-called optimum contrast A between
two Stokes matrices Ma and M_,, which is detined by
A - *T _la.ssT Mb,_ ( 2 )
where the polarisation slate of Stokes veclor ._ is chosen such that k is an extremum. In this method
first the ensemble averages of the Stokes lnatrices for all crop-type classes are calculated. Then for each
field we calculate the optimum ¢onl]'ast with all ('rop-tyI)e classes and the fit,hi is assigned to lhat class
for which the optimum ('ontrasl is a minimum.
2.3 Results
Single feature classificalion success percentages are between 2,0 and 50_(;. Generally tim cry.h results are
0 results. Th(, bes! results are obtained for the C- and L-ban(t cr_ on ,July' 3.better than the cr
The C- and L-band polarimetric success percentages are on average 60(7, for lhe I)CC method and
80% for the GML metho(l (see Fig. lay. The P-band results are significanlly lower for both methods,
since the backscatter of the soil dominates thal of the vegetation here.
When single features of the ,1 dates are combined into multi-temporal f('ature vectors success per-
centages of 70 to 80(7(, at'(, found (see Fig. 11)), so that it. ('an be concluded lhal single (lay polarimetric
classification is more l)owerful than 4 (lay' multiteml)oral classification in the mid to late growing season.
This situation may change when also data ohlaine(I in the beginning of the growing season is used.
3 Soil moisture determination of vegetated soil
For bare soil it is in l)rincil)le possible using radar to measure the top-layer soil moisture content if the
soil roughness is known, since lhe radar backscatter from soil 1)rilnarily (lel)ends on the soil moisture
content and soil roughness, ttowever, when the soil is vegetate(t we have 1o know the transmissivity of
the vegetation layer and also lhe relative ('ontributions in the hackscatter of lhe vegetation and of the
soil. This information cannot be obtained from single frequency and single l)olarisation systems (e.g. the
ERS-1 ), but maybe obtained from the three-frequency l)olarimetric AIRSAR system. Every measurement
with this system delivers 15 features (.5 features for each frequency band, see Secl. 2), which are certainly
not all independent, so lhat the dimensionality of the data-set is less than 15. If the dimensionality of
the data-set remains high enough, however, the information can possibly be use(] to solve for the different
contributions in the backscatter of vegeta|ed soil.
78
3.1 Description of the method
In orderto extractthe differentcontributionsin tile backscatterfrom vegetatedsoilweadoptthesimplemodelof Freemanand Durden(1992,hereafterthe FD model).This modeltransformsthepolarimetric
o o GOinformation (Ohh,avv, hv,P) into backscattering coefficients for diffuse, odd- and even-bounce scattering,which are related to tile interaction of the microwave radiation with the vegetation, with the ground and
with both the vegetation and the ground, respectively. Here we assume this is true for at least the C- and
L-band. For the P-band the diffuse scattering is certainly also affected by the soil. The diffuse scattering
in the model is estimated by assuming that the scatterers in the vegetation medium can be represented
by uniformly oriented and distributed small dipoles (needles). The ratio of the cross- and co-polarised
backscattering coefficient, which we call here the vegetation structure parameter r, is in this case 1/3.
The derived backscattering coefficient for the vegetation is directly related to this parameter.
The derived backscattering coefficient for the soil is related to the true backscattering coefficient by
a°,_,e = T] o o a',f,O) (3)cT,oi_(nh,, a', f, O) = e-'_f a,o_t( m_,,
where T! is the two-way transmissivity of the vegetation layer depending on the frequency f, a' the
soil roughness, 0 the incidence angle and m_ the volumetric soil moisture content. If we assume simi-
larly as in the FI) model that the vegetation backscattering is due to uuiformly oriented small needles
the transmissivity is described by Tj=e -'_j. If we have in addition an accurate model describing the
backscattering coefficients of the soil as a function of the depending parameters, and if the soil roughness
o' is known Eq. (3) contains only two unknowns a. and in v. In that case we can solve for a and my, once
the C- and L-band contributions of the soil are known from the FD model 1 As soil model we use the
empirical model of Oh el. al. (1992) which is valid for incidence angles > 20 ° and fi'equencies > 1 Ghz.
During the campaign soil roughness measurements were performed for some agricultural fields with
different crop types in Flevoland which are however generally valid since the soil composition and cul-
tivation are quite homogeneous in l:'levoland. We also obtained soil moisture measurements of a small
number of fields for the princit)al crot) types (potatoes, wheat, sugar beet and maize) in a part of the
observed agricultural area. Since the soil for potatoes is cultivated in furrows and ridges, which is not
described by the model of Oh et al. and the scattering by wheat and maize is often dominated by even
bounce scattering (especially in the L-band), we choose to use sugar beet in this study. We found 22
sugar beet fields in the selected area, which were vegetated during the three July overpasses.
3.2 Results
The soil roughness c,' is estimated to be 1.2 cm. (Vissers and van der Sanden, 1993). Unfortunately, the
uncertainty is rather large. Using this value for a' we solved for in, and cr requiring that the residue is
less than 0.1 dB in both the ('- and L-band. In this way we obtained soh, tions for 7, 8 and 18 fields
for July 3, 12 and 28, respectively. In Fig. 2a we show histograms of m_, Tc and TL for July 28. The
average soil moisture content is 0.5 gcm -a and the average transmissivity is 0.45 and 0.80 in the C- and
L-band, respectively.The measured soil moisture content in three sugar beet fields is about 0.25 (Vissers and van der
Sanden, 1993) so that the derived value is too high, although values derived from radar measurement
may be somewhat higher due to the big water-rich roots of the sugar beet plant. Also the value for the
transmissivity in the C-band is rather high, since Bouman ( 1991 ) found that the vegetation layer of sugar
beet in the C-band is probably opaque, so that values less than 0.3 are expected for To.
It appears that the solutions are rather sensitive to the value of the soil roughness parameter a' and
to the vegetation structure parameter r. If we estimate this value from measurements in the C-band,
tA problem may be that the penetration depth is wavelength dependent (,,- A/3), so that different frequencies probe
different soil moisture contents. We assume here that these differences are small in this context.
79
o 0.2 o.4 o.e o.sm, (g cm-s)
0 0.2 0.4 0.6 0.8 1
T/
|0 [-* " n " " v " " " u " " ! ....
[
i:m
...i...
0 0.2 0.4 0.8 0.8
m, (g cm-s)0 0.2 0.4 0.8 0.8 I
T/
Figure 2: Histograms of mv,Tc and TL (July 28), for a'=1.2 cm, r = 1/3 (Fig. 2a, left) and c'=1.5 cm,r = 1/4 (Fig 2b, right).
assuming that tile contribution of the vegetation dominates that of the soil (Bouman, 1991 ) we find valuesof 0.2 - 0.3 for r, which is smaller than the value of 1/3 in the FD model. Clearly the structure of the
sugar beet vegetation is also of importance and cannot simply be represented by small needle scatterers.
If we now change tile values for the vegetation structure parameter r to 1/4 and the soil roughness a' to1.5 cm we obtain reasonable results for July 28 (see Fig. 2b).
For July 3 and 12 no solutions were obtained in most cases, since on average the C-band soil con-tribution in the FD model is enhanced COml)ared to the L-band soil contribulio]l. This situation can be
explained when the vegetation structure parameter is lower for tile C-band than for the L-band, which
would imply that in the period between .luly 12 and 28 a change in the slructure i)arameters has oc-
curred. Indeed, measurements with the Et{S-I in 1992 show adrop in backscattering for sugar beets
during this period, which is probably related to a change in the vegetation structure (llijckenberg, privatecommunication).
We conclude therefore that there is an addional free parameter in the FI) model, which depends on
the vegetation structure and probably also on the fl'equency. Vegetation models like MIMICS (Ulaby et
al., 1990) may help to determine this parameter. Furthermore we need to know the soil roughzmss quiteaccurately in order to apply lhis method successfully.
4 References
(1) Bouman B.A.M., 1991. Li_king X-balM Radar Backscattering and Optical Reflectance with CropGrowth Models. 1)11. D. Thesis, Univ. of Wageningen, The Netherlands
(2) Freeman A., I)urden S.. 1992, "Filling a Three-component Scattering Model to Polarimetric SAR
Data", Proceedings of lh_ Third ,,ll_mml JPL Airbornc Gcoscie_we Wor£_hop, JPL Publication 92-1,1, Vo]. 3, pp. 5(i-5,_
(3) Kong J.A. (Ed.), 1990, Polarim¢lric Rome/( S'r1_sin9 (PIER 3), ]"lsevier, New York, Pl). 351-368
(4) O11 Y., Sarabandi K., Ulaby F.T., 1992, "An l.'nll)iri<-al Model and an Inversion Techni<lue for Radar
Scattering from Bare Soil Surfaces", IISEE Tr. Geo. Rein. Sens., vol. 30, no. 2, pp. :170-381
Ulaby F.T., Saral)an(ti K., McDonald M.W., Dobson M.C., 1990, "Michigan Microwave Canot)yScattering nm(lel",Int. ,1. Retool(_ ,5"<_sil_g, vol. 11, no. 7. I)P- 1223-1253
van den Broek A.(:., J.S. (;root, 1993, "lntercalibration of Satellite and Airborne SAIl", acceptedfor pul)lication in ('almdian dourwal of Remote Sc_tsing
Vissers M.A.M., van der Sanden J.J., 1993, Groundtruth Collectio_ for the JPL-5"AR and ERS-I(Ttmpaign i7_ FlecoloiM, B(:RS rei>orl n<). 92-26.
(5)
(6)
(r)
8O
N95- 23959
RELATING P-BAND AIRSAR BACKSCA'VrER TO FOREST STAND PARAMETERS
Yong Wang, John M. Melack, and Frank W. Davis
Center for Remote Sensing and Environmental Optics
University of California at Santa Barbara, Santa Barbara, CA 93106.
Eric S. Kasischke and Norman L. Christensen, Jr.
School of the Environment, Duke University, Durham, NC 27706.
1. Introduction
As part of research on forest ecosystems, the Jet Propulsion Laboratory (JPL) and
collaborating research teams have conducted multi-season airborne synthetic aperture
radar (AIRSAR) experiments in three forest ecosystems including temperate pine forest
(Duke Forest, North Carolina), boreal forest (Bonanza Creek Experimental Forest,
Alaska), and northern mixed hardwood-conifer forest (Michigan Biological Station,
Michigan). The major research goals were to improve undcrstanding of the relationships
between radar backscatter and phenological variables (e.g. stand density, tree size, etc.),
to improve radar backscatter models of tree canopy properties, and to develop a radar-
based scheme for monitoring forest phenological changes.
In September 1989, AIRSAR backscatter data were acquired over the Duke Forest.
As the aboveground biomass of the loblolly pine forest stands at the Duke Forest
increased, the SAR backscatter at C-, L-, and P-bands i,_creased and saturated at different
biomass levels for the C-band, L-band, and P-band data (Dobson et al. 1992). Due to the
4-page-length limit, we only use the P-band backscatter data and ground measurements to
study the relationships between the backscatter and stand density, the backscatter and
mean trunk dbh (diameter at breast height) of trees in the stands, and the backscatter and
stand basal area.
2. Study area and forest stand data
The tree stands used in this study are located in the Duke University Forest, which
is located west of Durham, North Carolina (36 ° 00' N, and 79 ° 00' W). The Duke Forest
contains forest stands with a total area of 3400 hectare, one-quarter of which are pure
stands of loblolly pine, Pinus taeda L. These pine stands range in age from < 1 to > 100
years in age.
This forest has been the site of ongoing research focused on developing a better
understanding of the potential use of imaging radars for monitoring southern U.S. pine
forests. Akborne SAR data were collected over this forest in 1984 and 1989 (Kasischke
and Christensen 1990, Kasischke et al. 1993a), and satellite data have collected with both
the ERS-I and JERS-1 SARs since 1991 (Kasischke et at. 1993b). This site represents
one Terrestrial Ecology Supersite that will be imaged by SIR-C/X-SAR system in 1994
(Evans et al. 1993).
A total of 23 pine stands are used in this study. The densities of the pine trees in
these stands range between 200 and 1844 trees/hectare. The average dbh of tree trunks in
81
thestandsrangesbetween13.7 and 33.9 cm. The average tree height in the stands ranges
from 11.7 to 25.6 m, and the average canopy depth from 5.3 to 9.2 m. The ground sur-
face in the selected 23 stands is level, which minimizes topographic effects on the SARdata.
3. Results
3.1. JPL AIRSAR backscatter data
JPL AIRSAR data were acquired on 2 September 1989. The data were collected
between 11:30 and 14:30, local time. The AIRSAR data were processed and calibrated
by using 8 ft, (2.44 m) trihedral corner reflectors. The estimated calibration uncertainty
was + 2.0 dB for P-band (0.68 m wavelength) backscatter. The standard 4-look com-
pressed data with pixei spacing of 12.1 m (azimuth) and 6.7 m (slant range) were pro-vided by JPL. To compute the mean of SAR data for a stand, we located the stand on the
SAR imagery, and the largest possible window within the stand was extracted. For each
stand, at least 200 image pixels were averaged.
3.2. Stand density vs. P-band SAR backscatter
There is almost no relationship between the P-band backscatter and stand density
(Figure 1). Of the 23 stands, as the stand density increases, tree size parameters (e.g. dbh,
tree height, and canopy depth) vary irregularly (Kasischke 1992). Thus, the stand density
is not a good parameter to explain the variation in the SAR back_atter.
3.3. Stand mean dbh vs. P-band SAR backscatter
As the mean trunk dbh of trees in the stands increases, the P-HH and P-HV
backscatter increases. The P-HH and P-HV back_alter is _aturated at stand mean dbh >
25 cm (approximately) (Figure 2a, c). The observed increase in backscatter may be
attributed to the increase of tree sizes. There is almost no relationship between the P-HV
backscattcr and stand mean dbh (Figure 2b),
3.4. Stand basal area vs. P-band SAR backscatter
The P-HH and P-HV back_atter increases when the sland basal areas increase.
There are large variations in the HH and HV backscatter for a given stand basal area (Fig-
ure 3a, c). The P-VV backscatter show ahnost no trend as the stand bas,'d area varies
(Figure 3b).
4. Concluding remarks
For Ioblolly pine stands at the Duke Forest, there is almost no correlation between
the observed AIRSAR P-HH, P-HV, and P-VV backscatter and stand density, and no cor-
relation between the P-VV backscatter and stand mean dbh or stand basal area. The P-
HH and P-HV backscatter increases as the stand mean dbh or the stand basal area
increases. The complex behavior of observed P-band backscatter from the Ioblolly pine
stands shown in this study can not be explained by a single stand parameter (such as
stand density, stand mean dbh, and stand basal area). Ongoing studies on the backscatter
by using forthcoming spaceborne and airborne SAR data, particularly multi-frequency,
multi-angle, and multi-polarization data, and by using a theoretical canopy backscatter
model coupled with the collected ground measurements (W_mg ct al. 1993) should help
complete the picture.
82
5. References
Dobson, M. C., Ulaby, E T., Le Toan, T., Beaudoin, A., Kasischke, E. S., and Chris-
tensen, Jr., N. L., 1992, Dependence of radar backscatter on coniferous forest
biomass, IEEE Trans. on Geosci. and Remote Sensing, 30(2):412-415.
Evans, D., Elachi, C., Stofan, E. R,, Holt, B., Way, J., Kobrick, M., Ottl, H., Pampoloni,
P., Vogt, M., Wall, S. van Zyl, J. J., and Schier, M., 1993, The Shuttle ImagingRadar-C and X-SAR Mission, EOS Transactions, 74:145.
Kasischke, E. S. and Christensen, Jr., N. L., 1990, Connecting forest ecosystem andmicrowave backscatter models, Int. J. of Remote Sensing, 11:1277-1298.
Kasischke, E. S., 1992, Monitoring changes in aboveground biomass in loblolly pine
forests using multichannel synthetic aperture radar data, Ph.D. dissertation, The
University of Michigan, 190 pp.
Kasischke, E. S., Christensen, Jr., N. L., and Haney, E., 1993a, Modeling of geometric
properties of loblolly pine tree and stand characteristics for use in radar backscattermodels, IEEE Trans. on Geosci. and Remote Sensing, in press.
Kasischke, E. S., Christensen, Jr., N. L., Bourgeau-Chavez, L. L., and Haney, E., 1993b,
Observation on the sensitivity of ERS-I SAR image intensity to changes in
aboveground biomass in young loblolly pine forests, Int. J. of Remote Sensing, in
press.
Wang, Y., Kasischke, E. S., Davis, F. W., Melack, J. M., and Christensen, Jr., N. L., 1993,Evaluating the potential for retrieval of loblolly pine forest biomass using SAR data
and a canopy backscatter model, to be submitted to IEEE Trans. on Geosci. and
Remote Sensing.
-5"10
-6
-_ -8
"r -927.
-10
-11
I | | | I
t
• * * t
t
0 500 1000 1500 2000
(a). Stand density (trees/ha)
_" -10
-11
-_ -13_f -14
(I: -15<_ -16
I | | | I
¢=
t t * •
¢ .
• 9,* t • .
e
0 500 1000 1500 2000
(b). Stand density (trees/ha)
_" -14
3 -15
_ -16
_ -17> -18
_ -19co -2o
I ! | | |,
. _ °t
" i;
t
t
o 500 1000 1500 2000
(c). Stand density (trees/ha)
Figure 1. Stand density vs. P-
band SAR backscattet for
lobloily pine forests at the
Duke Forest
83
"-" -5,.n
-6
-8
-i- -9'1"
-10
_-11
_" -14"D
-15
_ -16
_ -17> -18"l-
u) -20
• i | i
t
_ o m • ti•o°_
I
m
tm
10 15 20 25 30 35 40
(a). Stand _ dbh (cm)
! | |
t t
t t
to
t
_tt
10 15 20 25 30 35 40
(c). Stand mean dbh (o'n)
-10
_i -11
-12-13
_ -14
_ -15-16
• t
t •
ot it o Q_
t • t •
10 15 20 25 30 35 40
(b). Stand mean dbh (cm)
Figure 2. Stand mean dbh vs.
P-band SAR backscatter for
loblolly pine forests at the
Duke Forest
_" -5
-r -9
a: -10
m -1 1
_" -14'lO
-15
I -16-17
> -18
-19
_-20
| i i t i A
t
a
t
t_t •t
Q t Q_
w t,
I
. _ -11
-12
_ -13
_ -14n- -15
o9 -16
0 4O 80 120
(a). Stand basal area (m^?Jha)
t
tt_
• m •
I
t * t
• .i
0 40 80 120
(c). Stand basal area (m^2/ha)
| | a • • • •
t Q
• t
• t
t t t_,_
t
t m tt
0 40 80 120
(b). Stand basal area (m^2/ha)
Figure 3. Stand basal areas vs.
P-band SAR backscatter for
lobloUy pine forests at the
Duke Forest
84
N95- 23960
SOIL MOISTURE RETRIEVAL IN THE
OBERPFAFFENHOFEN TESTSITE USING
MAC EUROPE AIRSAR DATA
Tobias Wever & Jochen Henkel
Department for General and Applied GeologyWorking Group for Remote Sensing
80333 Munich, Germany
1. INTRODUCTION
Soil moisture content is an important parameter in many disciplines of science
like hydrology, meteorology, agriculture and others. Microwave remote sensingtechnique has a high potential in measuring the dielectric constant of soils, which is
strongly governed by the soil moisture (Ulaby et at. 1982). Much excellent work hasbeen done on investigating the relationship between backscattering coefficient and
soil moisture (Schmugge et al. 1980, Ulaby et at. 1986, Dobson et al. 1985, Rao etal. 1992, Shi et al. 1992). Most of these studies are measured in a laboratory or are
carried out with a multitemporal data set. This means, that the variation in thebackscattering coefficient is only related to the soil moisture because all other
parameters influencing the backscattering like surface roughness, vegetation cover,
plant geometry, phenology of plants and row direction are kept constant. In thisstudy the sensitivity of the backscattering coefficient to soil moisture of corn fieldsis investigated. In the framework of the MAC-Europe Campaign in June 1991, theNASA/JPL three-frequency polarimetric AIRSAR system collected data over the test
site Oberpfaffenhofen (Germany). The AIRSAR campaign in Oberpfaffenhofen was
complemented with intensive ground truth measurements. The sampled corn fields
are nearly in the range of the same incidence angle (=20") and belong to different
soil types. The evaluation was carried out at a single data set. The results show, that
the backscattering, measured at P-band can be described with only two parametersvery well. The main parameter, influencing the backscattering is the soil moisturecontent, the second subordinated parameter is the row direction.
2. ANALYSES AND RESULTS
For the assessment of the AIRSAR data for soil moisture retrieval all
frequencies (C-, L- and P-band) were investigated. Also different polarizations and
processing steps were used. Muitiparameter least square regression analysis wascarried out to fit the values of soil moisture (gray. % and vol. %) with these (er °) of
the AIRSAR system. The objective was to identify the best frequency and
polarization respectively the best processing steps for soil moisture retrieval overcorn fields. 17 corn fields with different SMC and different row directions relative to
the look direction have been sampled. All fields are placed within a range with a
similar incidence angle near 20 degrees to avoid effects referring to the incidence
angle. To get different SMC's, the cross section of sampled fields covers three
different types of soils: a loess soil, a waste gravel soil upon glacial gravel terracesand a drained ground-water soil. One problem is the spatial registration of the SMC
on the ground. Due to the fact, that only point measurements of SMC are possible,
the accuracy of the ground acquisition of SMC depends on the quantity ofmeasurements.
85
ThebackscatteringcoefficientatC-bandis mainly influenced by the interaction
between the incident wave and the vegetation cover, which is expressed by the goodperceptibility of the land use. The row direction takes no measurable effects and the
differences in the soil moisture are masked from the vegetation cover.
The same investigations were done for the L-band at all polarization combinations.Also at this frequency the backscattering coefficient is mainly influenced by thedifferences in the vegetation cover. However the canopy loss is smaller, which is
expressed by the week perceptibility of soil boundaries representing differences in
soil moisture content. The same is valid for the row direction tracing weekly onfields with the same vegetation cover, in this case corn.
Summarized it can be said that at L- and especially at C-band and an incidence
angle of approximate 20 ° the attenuation coefficient of vegetation canopies is tohigh for monitoring soil moisture without modelling.
Our measurements with P-band look very promising. The attenuation coefficient ofvegetation canopy is relative small, because now the soil moisture is the dominant
part influencing the backscattering (fig. 2). The second subordinated parameter isthe row direction (tab. 1). A multiparameter least square regression analysis with
the parameters row direction, soil moisture content and the backscatteringcoefficient was carried out. Figure 1 shows, that 92% of the backscattering can bede_ribed with the parameter row direction and gravimetric SMC at HH-HV
polarization and with HH polarization 90% of the backscattering. Table 1 illustratesthe composition of the total backscattering referring to the difference of HH-HVpolarization. It can be seen, that the main pan of the signal can be counted back to
the SMC. To get the single least square fit for gravimetric SMC the fields have to becalibrated to one fixed row direction (in this case 45°). Figure 4 shows the linear
least square fit between gravimetric SMC and o ° at HH-HV polarization. Thecorrelation coefficient of the relative to the row direction corrected data amounts to
0.83 at HH- respectively 0.85 at HH-HV polarization. Figure 3 demonstrates the
linear dependence of row direction and the backscattering coefficient.A comparison of the correlation coefficients illustrates the improvement taking a
nonlinear fit. The decreasing slope in the region of lower SMC concerning thenonlinear fit might be derived from the effects of bound water (fig. 5, tab. 2).
Fairly extensive studies demonstrate that the volumetric SMC represents thedielectric properties of different soils better than the gravimetric SMC (e.g. Scott &
Smith 1992). This study apparently do not confirm this thesis, but this effect mightbe derived from the little number of sampled fields referring to a statistical approachand the greater inaccuracy in measurement of the volumetric SMC on the grounddue to the small sample volume (100 cm 3) we have used (fig. 6).
The dotted lines in the figures are representing the average standard deviation of theground measurements.
A approach was carried out to detect the row direction automatically. With the
correlation coefficient between the HH and VV polarization there is a parameter forthe assessment of the row direction. For corn fields the influence of the row
direction is one order higher at VV polarization as for other polarizations. This is
caused by the strong interaction between the incidence wave and the vertical cornstalks. These interactions depend on the row direction, because the spacing between
corn plants within a row is much smaller than the spacing between the rows of corn.At HH polarization the influence of row direction is smaller and shows a more
linear dependance. These results are corresponding with a similar study of Brunfeldt& Ulaby (1984). The regression between the correlation coeffizients of HH- and VVpolarization and the row direction for each field was calculated. The correlation
coefficient of this regression amounts to 0.67. Using this regression line the anglebetween the row direction and the look direction can be assessed. A multiparameterleast square regression analysis was carried out taking the parameters calculated row
direction, grav. SMC ,and the backscattering coefficient of HH-HV polarization. Thecorrelation coefficient amounts to 0.79 which means that about 80% of the
86
backscatteringcanbedescribedwithoutcomplexmodellingofthevegetationcoveronlybythedataitself.Takingintoaccountarelativegreatinsecurityrepresentedbythemeasurementsof theSMConthegroundtheseresultssuggestthatforsoilmoistureretrievalfromSARimageryP-bandisverysuitableusingHH-orHH-HVpolarization.Withthetwoparameterssoilmoistureandknownrowdirectionabout90%respectively80%ofthetotalbackscatteringcanbedescribeddependingontheuseofgravimetric-orvolumetricSMC.Furtherinvestigationshavetobedonetoconfirmtheseresultswithotherplantcanopies.Supposingthattheinfluenceofrowdirectionisveryhighatcornduetothemarkeddifferencesinspacingbetweencornplantswithinarowandthespacingbetweentherowsofcorn,soilmoistureretrievalwithP-bandispromisingevenbetterresultsaboutotherlandusecanopies,especiallyatsmallerincidenceanglesbecausetheinfluenceofrowdirectionbecomessmaller.
3. REFERENCES
Brunfeidt D. R. & F. T. Ulaby (1984), "The Effect of Row Direction on the Microwave Emission from
Vegetation Canopies," IGARSS Symposium, Strasbourg, 27-30 Aug.Dobson M. C., F. T. Ulaby, M. T. Hallikainen and M. A. El Rayes (1985), "Microwave Dielectric
Bahavior of Wet Soils - Part II: Dielectric Mixing Models," IEEE Transactions on Geoscience and
Remote Sensing, Vol. GE-23, No. 1.Rao K. S., W. L. Teng and J. R. Wang (1992), "Processing of Polarimetric SAR Data for Soil Moisture
Estimation over Mahantango Watershed Area," Summaries of the Third Annual Geoscience Workshop,JPL Publication 92-14, Vol. 3.
Schmugge T. J., T. J. Jackson and H. L. McKim (1980), "Survey of Methods for Soil MoistureDetermination," Water Resources Research, Vol. 16, No. 6.
Scott W. R. & G. S. Smith (1992), "Measured Electrical Constitutive Parameters of Soils as Functions
of Frequency and Moisture Content," IEEE Transaction on Geoscience and Remote Sensing, Vol. 30,No. 3.
Shi J., J. J. van Zyl and E. T. Engman (1992), "Evaluation of Polarimetric SAR Parameters for SoilMoisture Retrieval," Summaries of the Third Annual Geoscience Workshop, JPL Publication 92-14,
Vol. 3.
Uiaby F. T.,R. K. Moore and A. K. Fung (1982), "Microwave Remote Sensing: Active and Passive,"Vol. II, Addison-Wesley, Readings, M. A.
Ulaby F. T.,R. K. Moore and A. K. Fung (1986), "Microwave Remote Sensing: Active and Passive,"Vol. III, Artech House, Norwood, M. A.
field
I
2
3
4
5
6
7
8
9
I0
11
12
13
14
15
16
17
Table 1.
row direction
[degmesl
35
23
23
15
45
66
75
75
75
4
8O
32
40
5
28
73
27
soil moisture
(gray. %)
17,34
14.91
13,61
15,83
15,66
16,39
20.54
20,27
Influence oi
soil moisture [dB]
7.70
6,62
6,04
7,03
6,96
7,28
9,12
9,DO
influence of
row direction [dB)
2.05
1,35
1,35
0,88
2,64
3,87
4,40
4.40
20.54
20,57
21,09
15.93
22.3
26,78
21,81
27,58
23,11
9,12 4,40
9,14 0,23
9,37 4,69
7,08 1,88
9,90 2,35
11,89 0,29
9,69 1,64
12,25 4,28
10.26 1,58
Composition of the total backscattering at
residue {dS]
0,81
0,83
0,65
0.61
1,20
0,45
1,57
0,26
0,08
0,80
1,38
2.39
0,01
1,59
1,03
0,19
1.36
HH-HV polarization.
87
-10
_k -16
-18
Fig. 1. Multiple linear regression for HH-HVpolarization.
-I1
-14
-16
-18
.-5
- I0
!,,-20 _ "
-25
I0 15 20 25
soil moisture (gray. 41
Fig. 2. Linear regression for HH-HVpolarization.
30
0
-5
-I015
-20
-25
20 40 60
row direction (deglee|)
8O
0
-5
,-10
-15
-20
-25
10 15 20 25
soil molstule (gray. % )
30
Fig. 3. Linear regression of row direction rela-tive to the look direction versus _r° at
HH-HV polarization after calibration toone fixed SMC.
-5
-10
-15
-20
12 14 16 18 20 22 24 26 28
soil rno_ure ray %
Fig. 5. Nonlinear regres, of grav. SMC versus
G° at HH polarization after calibration
to one fixed row direction (45°).
Fig. 4. Linear regression of grav. SMC versus
G° at HH-HV polarization after calib-
ration to one fixed row direction (45°).
0
-10 . _ • ...._ . . - -0" -
-15 .... -t; _.._-_D.. - -
"N -20 .....
-2515 2 2 30 35
loll molllure {vol. %)
Fig. 6. Linear regres, of vol. SMC versus
G° at HH-HV polarization after calib-
ration to one fixed row direction (45°).
data t_P-band)
gray. %
III! 0.73
HH-HV 0.73
IIH (calibrated) 0,83
HH-tIV 0.85
(calihrated)
Table 2.
linear regression
vol. % row direction
0.65 0.64
068 0.66
0.64 0.77
0.69 0.80
nonlinear regression
gray. %
x
x
0.89
0.87
mnltible linear regre_,fion
gray,% / row d,
0.9O
(I,92
x
x
vol.% / mw d.
0.79
0.82
x
x
gray.% / calc. row d
(J.g0
(I,79
×
x
Correlation coefficients of the linear and nonlinear regression analyses.
88
N95- 23961
MICROWAVE DIELECTRIC PROPERTIES OF BOREAL FOREST TREES
G. Xu, F. Ahem
Canada Centre for Remote Sensing
588 Booth Street, Ottawa, Canada K1A 0Y7J.Brown
Dendron Resource Survey Ltd.
880 Lady Ellen Place, Suite 206, Ottawa, Canada K1Z 5L9
INTRODUCTION
The knowledge of vegetation dielectric behaviour is important in studying the scattering
properties of the vegetation canopy and radar backscatter modelling. Until now, a limited number of
studies (Dobson et al., 1989, Hess et al., 1990, McDonald et al., 1992, Lang et al., 1993) have been
published on the dielectric properties in the boreal forest context. This paper presents the results of the
dielectric constant as a function of depth in the trunks of two common boreal forest species: black
spruce and trembling aspen, obtained from field measurements. The microwave penetration depth for
the two species is estimated at C, L and P bands and used to derive the equivalent dielectric constantfor the trunk as a whole. The backscatter modelling is carried out in the case of black spruce and the
results are compared with the JPL AIRSAR data. The sensitivity of the backscatter coefficient to thedielectric constant is also examined.
DIELECTRIC MEASUREMENT AND OBSERVATION
The dielectric measurements of the black spruce and trembling aspen were carried out during
the June of 1990 at the experimental forests of Petawawa National Forest Institute (PNFI), Ontario,
Canada. The experiment was part of the SAR-boreal project initiated at the Canada Centre for Remote
Sensing in 1988 in preparation for operational RADARSAT applications in forestry. The portable
dielectric probe (PDP) of Applied Microwave was used for these measurements. The dielectricmeasurements were made on the standing trees by incrementlly boring into the trunk and obtaining the
samples of the relative dielectric constant at each depth. The measurement proceeds along the radialaxis from the outer bark to the centre of tree trunks at the breast height of the tree (1.3 meters from
ground).
Figures 1 shows the profiles of real part of the relative dielectric constant inside the trunk of
black spruce and aspen at C, L and P bands respectively. The imaginary part is found to have high
linear correlation with the real part of the dielectric constant in general and is not illustrated here due to
the space limitation. It is seen that the dielectric constant peaks near the cambium layer and is much
lower in both the outer bark and sapwood regions. The heartwood region of the aspen displays a
higher dielectric constant than in the sapwood region, which is consistent with the field observation that
the heartwood region of aspen is often found to contain significantly more moisture content than in the
sapwood region. The dielectric profiles fluctuate, reflecting the inhomogeneity of the medium.
PENETRATION DEPTH AND EQUIVALENT DIELECTRIC CONSTANT
Depending on the microwave penetration depth and the physical size of the tree componentssuch as the branches and trunks, the microwave interacts with the portion or the whole of these
vegetation materials. It is therefore necessary to consider the microwave penetration depth at different
frequencies and incorporate it into the derivation of the equivalent dielectric constant of different tree
components for the radar backscatter modelling.
The microwave penetration depth is estimated using the formula (Ulaby et al., 1986, Pg.2019):
1 c
89
where _ is the absorption constant of the medium, c is the speed of the light, f is the microwave
frequency, _' and _" are the real and imaginary part of the equivalent dielectric constant of the medium.
As a first approximation to the equivalent dielectric constant, the measured trunk moisture
content of trunk wood samples of two species together with the corresponding dry density (Gonzalez,
1990) are input to the dual dispersion dielectric model developed by Ulaby and EI-Rayes (Ulaby et al.,1987) to derive the equivalent dielectric constant for these species. Table 1 lists the calculated
dielectric constant at C, L and P bands. These dielectric constants are then substituted into the Eqn. 1to obtain the estimated microwave penetration depth, as is shown in Table 1.
Since the power of an electromagnetic plane wave inside a lossy medium decays at a rate of e _"'where r is the distance from the surface of the medium, we compute the equivalent dielectric constant
of the tree trunk with the microwave power decay rate as a weighing factor, that is:
< _> _ f¢(,)_-'o'a, (2)f e-Z,,,dr
where the integral is taken from the bark surface to the centre of the trunk. Applying Eqn.2 to thesampled dielectric profile, we obtain the equivalent weighted dielectric constant. This is illustrated inTable 1.
BACKSCATTER MODELLING AND SENSITIVITY ANALYSIS
An example is given here to illustrate the effect on the radar backscatter with the dielectric
constant derived from two different approaches, namely the dielectric model and the PDP sampled data.The Michigan Microwave Canopy Scattering (MIMICS) model (Ulaby et al., 1990) is applied to atypical black spruce stand in the Canadian boreal forest environment. Table 2 shows the backscatter
coefficients predicted from the MIMICS model and obtained from the JPL AIRSAR data which were
acquired in May, 1991 over the boreal forest test site near Whitecourt, Alberta. Despite the different
approaches in deriving the dielectric constant, the results show a very good agreement in the predictedbackscatter coefficient, with a discrepancy within one dB for like polarization and less than two dB for
the cross polarization for all three microwave frequency. The predicted backscatter coeffcients agreeexceptionally well with the SAR observation at C band, in particular with the dielectric constant derived
based on the microwave penetration depth. The MIMICS model, however, is seen to over-predict thebackscatter coefficient for like polarization at L band and under-predict the backscatter coeffcient for
cross polarization at L and P bands, which is presumably caused by the fact that the MIMICS model
does not account for the multiple (triple and over) scattering mechanisms in the computation of thebackscatter coefficient.
Table 3 illustrate the effect of the dielectric constant on the backscatter coefficient. The
dielectric constant of the black spruce is perturbed by 10 percent and the corresponding change in thepredicted backscatter coefficient is calculated. It shows that the backscatter is more sensitive to the
change in dielectric constant for like polarization at C band than at L and P bands. Among two likepolarizations at C band, the backscatter is more sensitive to the change in dielectric constant for HHpolarization than for VV polarization.
ACKNOWLEDGMENTS
The authors wish to thank I.MacRae, J.Drieman, C.Skelly, D.Leckie, S.Yatabe and B.Rice forassisting the dielectric measurements.
REFERENCES
Dobson, M.C., McDonald, K., Ulaby, F.T. 1989. Modeling of Forest Canopies and Analysis of Polarimetric
90
SAR data. ERIM Contract Report, No:211000-1
Gonzalez, J.S. 1990. Wood Density of Canadian Tree Species. Information Report NOR-X-315, Northern
Forestry Centre, Forestry Canada
Hess, L. and Simonett, S. 1990. Dielectric Properties of Two Western Coniferous Tree Species. Proe. of
IGARSS' 90 Symposium, p.503
Lang, R.H., Landry, R., Kilic, O., Chauhan, N., Khadr, N., Leckie, D. 1993. Effect of Species Structure and
Dielectric Constant on C-band Forest Backscatter. Proe. of IGARSS'93 Symposium
McDonald, K.C., Zimmermann, R., Way, J. 1992. An Investigation of The Relationship Between Tree Water
Potential and Dielectric Constant. Pros. of IGARSS' 92 Symposimn, pp.523-525
Ulaby, F.T., Moore, R.K., Fung, A.K. 1986. Microwave Re,note Sensing, Vol.lll. Arteeh House, inc.
Ulaby, F.T., EI-Rayes, M. 1987. Microwave Dielectric Spectrum of Vegetation-Part II: Dual-Dispersion Model.
IEEE Trans. on Geoscience and Remote Sensing, VoI.GE-25,5:550-557
Ulaby, F.T., Sarabandi, K., McDonald, K., Whitt, M., Dobson, M. 1990. Michigan Canopy Scattering Model.
Int. J. Remote Sensing, Vol.ll, 7:1223-1253
Band
J SpruceM_.=0.31 tr[ D.:0.46
Penetration Depth=Q._ cm
tl=10.65-j3.47 C_=17.14- _6.06
Ampen
'Mq.-0.43 Dry D.=0.42
Penetration Depth=O._l 7m
e1=15.50-js.03 C:=14.42-j5.88
Penetration Depth=3.06 cm Penetration Depth-2.63 cmL
_i =13. 31-j4.59 _=14.05-j4.10
Penetration Depth=6.08 cm
c.=15.80-j7.10 _z=15. i0 -j3.6 ,I
_=18.88-j6.35 _2=14.7"1-j3.78
Penetration Depth=4.89 cm
C1=21.89-j10.4 ez=17.71-j4.75
Table i. The ettimated microwave penetration depth and equivalent dielectric
constant of black spruce and trembling aspen tree trunks.
Mg.: the gravimetrlc moisture content (%)
Dry D.: dry denlity (9/cm'3)
_i: the equivalent trunk dielectric constant predxcted from the dielmctric model
82: the equivalent trunk dielectric constant derived from the PDP measure_4nt
Band
C-HH
o°(c:) _o°(+lO%z,)
[ -9.38 ] -0.71
4o°(-I0%¢ I)
0.88
C-W -8.79 J -0.32 (I. 39
C-HV -16.46 O. 01 -0.06
L-HH -2 . 49 -0.17 0.14
L-W -8.69 0.09 -0.14
L-HV -24.48 0.82 -0.91
P-HH -5.43 0.29 -0.16
P-W -13.39 0.37 -0.24
P-HV -28.87 0.31
The backsc_tter coefficient (in dB) Table 3.
predicted with MIMICS model and
obtained from AIRSAR obmervetion
for black spruce stand.
-0.19
The sensitivity of backacatter coefficient tothe Change in dielectric constant. The diels-
cilia constant le _ubed 10 percent and thecorresponding back_tter change (in dUB) isshot_.
91 ORIGINAL PAGE IS
OF POOR QUALITY
r
" t0
Black Spruce
C
_s
4', +
J5 t
,+;i _Voo ,+F
o + --- --+ ....
0
Trembling Aspen
C
L
,,_++
,+_ t.
IS
i,,+,: + .....
i
o J, ...... +÷----
0 I00
P
/
a 188 m
p
_o. +-;
+Olin 11oo 141t1 o _ 411o wo Ill
"t:_:
I 111 lelO
Probe Depth (I00 = 1 cm) Probe Depth (I00 = 1 cm)
Figure I. Dielectric profiles (real part) of black spruce and
trembling aspen tree trunks. The vertical scale is
the real part of PDP sampled dielectric constant.
92
Slide No.
10
11
12
SLIDE CAPTIONS
Caption
3-component RGB image of the study area; AVIPdS
data collected in August 1990. Blue = AVIRIS band
12; green = band 48; red = band 213.
Jasper Ridge example: The five endmember spectra
and their corresponding abundance images.
Color composites of the simulated and actual TMscenes.
Selected linear spectral unmixing results for both
the empirical line and atmospheric-modelcalibrated data.
Classification map, produced by a self-organizingANN.
A two-band image showing the spatial
relationship between the derived lignin and waterfractions across the site.
(A) shows the composite rendered on the digital
elevation model with tags (B-E) for the location of
the enlarged subareas of the horizontal (non-
rendered) composite. (B)-(E) show zoom-up
(=factor 5) parts with different aspect and slope
angles.
Fractional abundance images and associated
spectra of semi-arid landscape endmembers,
August 7, 1990.
AVIRIS image cube of Rogers Dry Lake, CA,calibration site.
AVIRIS image cube of Moffett Field, CA, data set
showing San Francisco Bay and Moffett Field
runways.
The classified June 2nd image next to a vegetation
map (mapped in the field) for comparison.
AIRSAR views of Aeolian terrain.
SeniorAuthor
Beratan
Boardman
Kalman
Kruse
Mer6nyi
Wessman
Meyer
Yuhas
Green
Green
Sabol, Jr.
Blumberg
93